|
DIGITAL EQUIPMENT INTRODUCES FAMILY OF
MASSIVELY PARALLEL COMPUTERS
MAYNARD, Mass. -- August 20, 1991 -- Digital Equipment Corporation
today introduced a family of computers based on massively parallel
processing (MPP), an emerging, high performance technology targeted
at very complex problems of technical, scientific and commercial
users.
The new DECmpp 12000 system series features configurations
ranging from 1,024 to 16,384 processors with performance up to 1.2
billions of floating point operations per second (GigaFLOPS), and
26,000 millions of instructions per second (MIPS).
The new systems, supplied under a previously announced
agreement with MasPar Computer Corporation of Sunnyvale, California
are priced from $240,500 to $1,502,500. Worldwide shipments will
begin in the fall.
All DECmpp systems are configured with Digital DECstation 5000
full color front-end workstations, as well as ULTRIX and DECnet
software licenses. In addition to the family of computer
processors, Digital also announced a comprehensive line of
software, a family of high speed disk arrays, and full service and
support programs. Digital's standard one year product warranty
will apply.
With this announcement, Digital becomes the first main line
systems vendor to formally introduce a family of massively parallel
systems, and it is the first time users of systems in this
performance class can take advantage of Digital's high level of
warranty, service and support.
Charles Wilson, director of Digital's Massively Parallel
Systems Group, noted "We feel this is a milestone in the evolution
of massively parallel technology. There are many businesses,
government agencies and universities that want to move to massively
parallel technology, but are looking to more application software
and higher levels of enterprise service and support.
"Digital's entry in the market is important to these users for
two reasons. First, we will stimulate application program
development both directly through our independent software
developers, and indirectly by adding the commitment of an
established, major systems vendor to this relatively new market.
"The second reason is Digital's leadership in delivering
effective, highly reliable, enterprise-wide support to large
organizations. MPP systems are yet another important technology to
be integrated into the total computing environment."
Introduced today were:
o DECmpp 12000 systems -- eight field-upgradable models
incorporating 1,024, 4,096, 8,192 and 16,384 processor
elements, and memory capacity from 16MB to 1GB. Priced from
$240,500 to $1,502,500;
o DECmpp Programming Language -- a C-like high level compiler
for programming DECmpp systems included with the DECmpp
System Software license, $100;
o DECmpp Parallel Programming Environment -- a comprehensive
suite of development tools for data-parallel programming.
Features an interactive DECwindows point-and-select
interface. Includes symbolic debugger, machine animator and
data visualizer. License, $4,000;
o DECmpp Image Processing Library -- a facility for supplying
common visual and data objects to help speed program
development. License, $4,000.
o DECmpp FORTRAN -- built on FORTRAN 90, this compiler will
provide facilities to program in high level language and
generate code for data parallel execution. Available in
December;
o DECmpp Disk Array Systems -- high speed disk-array
subsystems, provide parallel access to large data files.
The disk array features 720 megabytes of formatted storage
per disk with a capacity up to 11.2 gigabytes of storage
and sustained I/O rate of nine megabytes per second.
Priced from $129,000 to $261,000;
o Digital Service -- the complete worldwide service offering
from Digital, including enterprise integration, consulting,
customer training, and maintenance and repair services, is
available to DECmpp customers.
SOFTWARE AIDS MARKET GROWTH
Massively parallel processor architectures are considered by
many industry experts to have the greatest potential for solving
highly complex problems in technical, scientific and commercial
computer applications such as scientific visualization, geographic
database management, econometric modeling, and financial analysis.
MPP computers are able to break down problems into smaller
parts, then process these parts simultaneously, at very high speed.
As a result, performance can be extremely fast for applications
that lend themselves to this type of processing.
Key to market acceptance, however, is software, ranging from
application development and network integration tools to the
applications themselves that can take advantage of this technique.
Digital is working to advance MPP software development on
several fronts, according to Wilson.
"First, we're focusing on basic software elements such as
language compilers, libraries, networking designs and basic
development tools. In some cases, we're going to create
parallelized versions of existing products; in other cases, we're
developing entire new products. For example, we are looking to
extend Digital tools such as our Language Sensitive Editor,
Friendly UNIX, and Test Manager to massively parallel
architectures.
"Secondly, we're committed to directly stimulating third-party
application development in strategic application areas."
Targeted applications include: structural dynamics in
automotive and aerospace industries; seismic processing and
reservoir analysis in oil and gas exploration; and database
retrieval, searching and sorting.
Two application toolkits recently ported to these systems are
"Elexir", an information retrieval system from Third Eye Software,
Inc. of Menlo Park, California, and "Table-Maker", a statistical
database package from CHISQ Ltd., Reading, England.
DECmpp EXTENDS HIGH PERFORMANCE SOLUTIONS
The new DECmpp 12000 Series Systems add an important MPP
component to Digital's high-performance computing products.
Digital currently manufactures and markets VAX 9000 and VAX 6000
systems, including mainframes and vector equipped versions for
supercomputing. Digital also offers a range of high-performance
workstations. These and other Digital products maintain industry
leading interoperability with non-Digital systems via the company's
open NAS architecture of standard software interfaces.
Digital Equipment Corporation, headquartered in Maynard,
Massachusetts, is the leading worldwide supplier of networked
computer systems, software and services. Digital pioneered and
leads the industry in interactive, distributed and multi-vendor
computing. Digital and its partners deliver the power to use the
best integrated solutions -- from desktop to data center -- in open
information environments.
####
CORP/92/284
Note to Editors: MasPar is a trademark of MasPar Computer
Corporation.
Table-Maker is a trademark of CHISQ Ltd.
Elexir is a registered trademark of Third Eye
Software, Inc.
UNIX is a trademark of Unix System Laboratories,
Inc.
DECnet, DECstation, DECmpp, VAX and ULTRIX are
trademarks of Digital Equipment Corporation.
CURRENT APPLICATIONS
The newer massively parallel architectures make it possible to
process computationally intensive applications with ultra high
speed and relatively low cost. For example, a 16,000-processor
machine is rated at 26,000 peak MIPS and 1,200 peak MFLOPS, and
costs just over $1 million.
Some typical applications are:
o Structural analysis -- used frequently in automotive and
aerospace industries, this relates to finite element
analysis of mechanical structures, both at rest and under
stress;
o Computational fluid dynamics -- the fastest growing area of
computational physics, this typically involves the study of
aerodynamics, including airflow over airfoils and air
resistance of automobiles;
o Seismic/reservoir modeling -- these are the two most used
oil-industry applications;
o Signal and image processing -- used extensively by NASA and
the military, this application requires very fast filtering
and analysis of electronic signals;
o Chemistry/biology -- these applications include studies of
chemical structures, reactions and properties, as well as
genetic molecular modeling;
o Commercial -- the major potential today for business/office
users involves database, financial-modeling and text-
retrieval applications.
|
| HIGHLIGHTS
o Among our competitors, Digital is the first major company to formally
enter the market for massively parallel computer systems
o Provides up to 16,384 processing elements capable of delivering 26,000
MIPS (32-bit add), 1.2 gigaflops 32-bit floating point
o DECstation 5000 front-end systems use industry standard ULTRIX operating
systems
o A robust software development environment through high-level language
data parallel programming
o Digital integrates MPP into a customer's distributed computing
environment
o Prices range from $264K -- $1.5M; delivery planned for September 1991
INTRODUCTION
Digital introduces a family of computers based on massively parallel
processing (MPP), an emerging, high performance technology targeted at the
very complex problems of technical, scientific and commercial users.
The new DECmpp 12000 system series features configurations ranging from
1,024 to 16,384 processors with performance up to 1.2 billions of floating
point operations per second (gigaflops), and 26,000 millions of instructions
per second (MIPS).
The new systems, supplied under a previously announced agreement with MasPar
Computer Corporation of Sunnyvale, California, are priced from $264,000 to
$1,502,500.
ONE CONSISTENT COMPUTING ENVIRONMENT
Digital now has a leadership position in high performance computing through
our DECmpp 12000 massively parallel processing systems, VAX 9000 systems,
networks and applications. We have the ability to integrate all of these in
one consistent computing environment for computationally intensive
applications.
Key to market acceptance, however, is software, ranging from application
development and network integration tools to the applications themselves,
that can take advantage of this technique.
Focus on Basic Software Elements
First, we're focusing on basic software elements such as language compilers,
libraries, networking designs and basic development tools. In some cases,
we're going to create parallelized versions of existing products; in other
cases, we're developing entire new products. For example, we are looking to
extend Digital tools such as our Language Sensitive Editor, Friendly UNIX,
and Test Manager to massively parallel architectures.
Stimulate Third-Party Applications
Secondly, we're committed to directly stimulating third-party application
development in strategic application areas.
Two application toolkits recently ported to these systems are "Elexir", an
information retrieval system from Third Eye Software, Inc. of Menlo Park,
California, and "Table-Maker", a statistical database package from CHISQ
Ltd., Reading, England.
Strategic Development Technologies
Digital is focusing its resources on strategic development technologies and
activities by:
o Developing and enhancing application development tools -- Digital will
extend development tool components of its CASE (Computer Aided Software
Engineering) product family to work with massively parallel
architectures; Digital will also pursue advance development in other key
technologies such as language compilers, data libraries and networking
technologies
o Stimulating third-party development activities -- Digital will pursue
partnerships and other initiatives to assist third-party application
developers with the creation of applications that take advantage of
massively parallel architectures
HOW MASSIVELY PARALLEL COMPUTERS WORK
Massively parallel processing is an outgrowth of the quest to speed computer
processing. The traditional computer architecture (known as the von Neumann
architecture) operates in serial fashion. With every cycle of a central
computer clock, the central processor unit (CPU) processes a fixed-size
instruction or word. The larger the word length, from 16 to 32 bits, for
instance, and the faster the processor, the greater the volume of
information that can be processed in a set time.
However, the faster the CPU, the more memory and larger input/output
bandwidth is required to maintain balanced, high-speed operations.
Multiple-processor implementations, with four, six, or even a hundred or
more processors, can bring significant increases in speed, but they must be
coordinated by additional logic in order to maintain synchronization. This
interprocessor overhead can add considerably to internal resource
utilization.
Massively parallel architectures employ thousands of relatively inexpensive
processor/memory units, distributed throughout the system. MPP
architectures are thus very efficient at handling large data arrays (of
1,000 by 1,000 elements, for instance), where the instruction flow can be
parallelized to operate on all elements simultaneously.
As a result, massively parallel computers are considered to have the
potential, as more application software becomes available, for handling a
broadening range of highly complex problem-solving applications.
PRODUCT DESCRIPTION
The DECmpp 12000 systems series consists of eight models, depending on the
number of Processor Element (PE) boards and memory per board. The DECmpp
12000 Series can be configured with 1024, 4096, 8192, or 16,384 processor
elements at initial installation or through on-site upgrades. By operating
in a Single Instruction Multiple Data (SIMD) fashion, the thousands of
processor elements can work on a single problem simultaneously. An entry
level DECmpp system provides 1,600 MIPS of processing power and is scalable
up to 26,000 MIPS -- and all within the same system package.
DECmpp Software Products
Although the DECmpp 12000 Series uses many interconnected processing
elements to achieve high performance, it features a single, integrated
software environment. This environment is based on the ULTRIX operating
system giving access to hundreds of utilities and applications. The user
interface to the software environment is based on the DECwindows system and
supports access from DECwindow terminals and DECstation workstations.
DECmpp Parallel Programming Environment
The DECmpp Parallel Programming Environment consists of a comprehensive set
of development tools for data parallel programming. This easy-to-use
environment features an interactive DECwindows-based point-and-select
interface. Users will easily visualize, analyze and optimize the behavior
of DECmpp processor elements, algorithms, and data. The DECmpp tool set
includes a symbolic debugger, a machine animator and data visualizer.
DECmpp Programming Language
DECmpp System software includes a C-like, high-level language compiler for
programming the DECmpp 12000 Series systems. The language is based on
Kernighan and Ritchie C and is the primary programming language for the
DECmpp system's data parallel unit.
DECmpp Image Processing Library
The DECmpp Image Processing Library is a highly optimized set of image
processing routines for the DECmpp 12000 Series Systems. The library
supports the development of very high performance image processing solutions
on a fully programmable platform with a symbolic, window-based, interactive,
graphical, and network-distributed development environment. Library
routines are callable from standard languages such as C, FORTRAN, and DECmpp
Programming Language. The library includes routines for image arithmetic,
low-level image processing operations, and image analysis, as well as a
variety of utilities for image display, I/O, test image generation, and
format conversion.
DECmpp Parallel Disk Array
The disk array provides the DECmpp 12000 Series systems with up to 11.2
gigabytes of formatted capacity and 9 MB/Sec sustained disk I/O. The disk
array is ideal for a wide range of data-intensive applications, such as
scientific visualization, remote sensing, image processing, and geographic
database management.
The disk array can be upgraded from four disks at 3 MB/Sec to 24 disks at 9
MB/Sec. Digital's TKZ08 tape drive may used to back up the parallel disk
array.
THE DIGITAL MPP STRATEGY
Digital has signed an OEM agreement with MasPar Computer Corporation to
resell MasPar's massively parallel computer systems, software and options.
The DECmpp 12000 systems combine excellent price/performance to provide a
highly productive growth path for the future. The product line is easily
scalable, upgradable and includes software tools for developing parallel
processing applications.
Digital's strategy is to:
o Provide an integrated environment by drawing on our strengths in areas
such as:
- Software development tools
- High performance networking
- System integration
- Workstations and servers
- Third-party software relationships
o Establish partnerships with leading MPP hardware and software companies
o Expand the market by raising awareness of the benefits of MPP
Why Did Digital Choose to Resell MasPar's Systems?
MasPar's products have high quality and an excellent price/performance. The
products are designed and built on state-of-the-art technology licensed from
Digital. This is the first partnership, but another, MIMD (Multiple
Instruction Multiple Data) partnership is planned for later this year.
PERFORMANCE POSITIONING
LINPACK
Peak 1K x 1K * Price
MIPS Number of MFLOPS 64-bit Range
32-bit Processors 64-bit MFLOPS $K
DECmpp 12000 26,000 16K 580 440 $ 274 -
Model 16B $1,533
Thinking 2,500 2K 14,000 5,200 $2,000 -
Machines CM-2 $7,000
Intel iPSC/860 1,056 32 1,300 486 $1,000 -
128 5,000 1,920 $4,000
VAX 9000 /410VP 30 1 125 79 $1,000 -
440VP 30 4 500 n/a $6,000
Cray Y-MP/132 N/A 1 333 324 $1,500 -
/832 8 2,667 2,144 $24,000
Convex C3210 N/A 1 50 17 $1,000
C3240 4 200 166 $6,000
C3810 1 106 99
* From report generated by Jack Dongarra, University of Tennessee, 6/12/91.
POSITIONING WITH RESPECT TO VAX 9000
The VAX 9000 vector processing system provides excellent performance in a
wide range of applications readily available from Independent Software
Vendors covering Science, Engineering, Chemical, Pharmaceutical and Oil and
Gas Industries. It also has excellent scalar and I/O performance, meaning
that it is a very balanced and versatile system to answer the growth needs
of the immense installed base of VAX system users in all segments. It will
very often be used as a compute or database server to satisfy the single
stream performance needs of an organization consisting of workstations or
time-sharing users on smaller VAX systems.
DECmpp 12000 Series system is a specialized execution engine capable of
extremely high performance for a more limited set of applications. Today,
it is relatively more difficult to program, but can yield dramatic execution
speed dividends on applications that involve computing on large arrays of
data. The DECmpp 12000 systems include a DECstation 5000/200, which
functions as a general purpose host computer to control and to execute
sequential code. VAX 9000 users can access the data parallel unit through
the DECstation workstation.
MARKETS AND TYPICAL APPLICATIONS
Markets for DECmpp systems include industries such as Government, Education,
Oil and Gas, Chemical, Pharmaceutical, Automotive/Aerospace, Service and
Financial, and Defense.
Typical applications include computational fluid dynamics, geophysical data
processing, structural dynamics and acoustics, ocean and atmospheric
modeling, computational chemistry and physics, simulation of neural nets,
molecular modeling, database and text retrieval, searching and sorting,
image processing and recognition, signal processing (1D, 2D, 3D), and
seismic processing and reservoir analysis.
The Massively Parallel Systems Group (MPSG) has an aggressive program to
provide off-the-shelf application software packages for DECmpp systems.
Please contact the MPSG staff for status and availability of application
software packages.
PRICING/ORDERING/CONFIGURATION/AVAILABILITY
All DECmpp 12000 Series systems include a DECstation 5000/200CX workstation
with an additional 16 MB RAM, VME adapter and SZ12B-FA sidecar, which
includes an RZ56 hard disk and a RRD42 CDROM drive, licenses for ULTRIX
workstation software and DECnet. Also included is a one-year Basic Service
warranty for all DECmpp 12000 Series systems.
The products listed below are also available from MasPar, but are priced
higher to represent added value when purchased from Digital. Digital's
prices include our comprehensive one-year warranty (versus MasPar's 90-day
return-to-factory warranty), our unlimited users per system software
licensing, our long-term commitment to support these products anywhere in
the world, and our investment in growing the value of these systems through
increasing application and system software.
|
| Below is the latest edition of the newsletter put out by MPSG
(Massively Parallel Software Group) in Maynard; this is the group
working with MASPAR on the development of software for the MASPAR
machines.
To subscribe, send mail to Darlene Camuti at RDVAX::CAMUTI.
Regards,
Mark
..............................................................
From: FASDER::RDVAX::NILSSON "Bob Nilsson, 223-1715, MPSG 23-Aug-1991 2218" 23-AUG-1991 22:57:25.21
To: @[CAMUTI]MPP.DIS
CC: NILSSON
Subj: Massively Parallel Systems Newsletter #4 - August
D I G I T A L
- ____ ____ ____ ____
- / / / /___/ /____ / __
- / / / / ____/ /____/
16,384 Points of Light
MPSG NEWSLETTER #4
Massively Parallel Systems Group - August Update
***Digital Internal Use Only***
Attached is the fourth update from MPSG on Digital's MPP program and
plans. Please let Darlene Camuti (RDVAX::CAMUTI) know if you would
like to be deleted from this distribution list or if you know of
someone who should be added.
Contents
--------
1. Current program status
2. Training
3. MasPar MPP installed sites
4. DECmpp notesfile, public directory
A1. DRAFT DECmpp Sales Update article (with prices and
MPSG contacts)
A2. DECmpp Press Release
NOTE: Please remember to call us at MPSG before you call
a MasPar sales rep. DECmpp is now a Digital product and our
selling success is maximized when we, Digital, present DECmpp
to the customer. It is often appropriate to bring MasPar
into our selling process, but only after we have presented
our products and strategy to the customer. Please see the
attached sales update article for a list of MPSG contacts to
call. As a Digital sales rep you will get full credit for
DECmpp systems which Digital sells to your customer.
1. Current Status
-----------------
This was a momentous week for DECmpp! The pricing was
approved on Monday and we thank PAC for their cooperation
during Hurricane Bob. We announced DECmpp on Tuesday and
despite some other newsworthy activity in the Kremlin, the
press pick up has been terrific! The NY Times, Wall Street
Journal, Boston Globe, Nightly Business Report among others
ran stories. Articles based on interviews with us are also
expected in: Computerworld, Computer Systems News,
Electronic News, EE Times, Federal Computer Week, Government
Computer News, Information Week, Digital News, Electronic
Business, Digital Review, Electronics, Datamation,
DEComputing, Supercomputing Review, Parallelogram,
Computing, Computer Weekly, Computergram.
This week Albuquerque manufacturing people built their first
DECmpp systems. The first US DECmpp sale was installed at
Georgia Tech and was up and running in less than 2.5 hours.
More DECmpp events:
August 23 Sales Flash #693 announces Digital MPP systems in price
list
September 16 Sales Update cover story on DECmpp
September 9-13 DECUS, The Hague. DECmpp demos and presentations
October 22 Cornell University DECmpp day (tentative)
November 13-15 DECUS, Japan. DECmpp demos and presentations
November 10-14 SEG show, Houston. DECmpp demos
November 14-16 IEEE Supercomputing Show, Albuquerque. DECmpp demos
December 9-13 DECUS, Anaheim. DECmpp demos and presentations
Planning is beginning for a Washington, DC area DECmpp 1-day
seminar. Steve Harrington will have more details in our next
newsletter.
2. Training
-----------
We are still targeting September 25-27 for sales and support
training in Massachusetts. Let Darlene Camuti know if you
and your manager feel that you should attend this training.
Sales and sales support individuals with DECmpp prospects
will be given priority for the limited seating.
We are looking to provide a two-week in-depth DECmpp
programming course in early October. Capacity for this will
be limited, so let Darlene Camuti know if you would like to
attend this October Systems Engineering training.
Additional training is also being planned for Europe. Call
Steve Zagame or Yvan Bantoure for more info (see contact
list in the attached Sales Update article.
Massachusetts MPSG Sales and Support Training Agenda
DAY 1
8:30 Introduction to MPSG and goals of training
9:00 MPP style of computing, Digital strategy and plans
10:30 break
10:45 MasPar background
11:00 DECmpp Products
11:30 lunch
12:30 MPP product training part 1
2:00 Break
2:15 MPP product training part 2
3:45 MPP positioning and Day 1 wrap up
4:00 Demos
5:15 End of day 1
6:30 Dinner with MasPar
DAY 2
8:30 MPP applications
9:00 Selling MPP systems vs the competition
10:00 break
10:30 Marketing events
10:45 Configuring, quoting, site prep, Digital MPP mfg
11:15 Customer support
12:00 Lunch
1:00 Hardware futures
1:30 Software futures
2:00 break
2:15 Description of text retrieval demo
2:45 Discussion and wrap up
DAY 3
1 x 1 consulting on customer opportunities
optional MPP programming instruction
3. MasPar Installed sites and applications
------------------------------------------
We have received many requests for information on installed
MPP sites. Since DECmpp is a new Digital product, there are
not yet many installed Digital sites. Below are some MasPar
sites to give you an idea of current applications for MPP.
However, please do not contact these customers directly. If
you are interested in more information, give us a call or
call the MasPar representative associated with the site.
Customer Application MasPar sales rep
-------- ----------- ----------------
Ford Aerospace Neural networks Ed McCurtain
UNC Neural Nets, computational Al Payne
chemistry, molecular modeling
Ames Laboratory SIMD research Gerry Winckler
computational chemistry
NASA Goddard galactic nebulae gas Will Workman
velocity modeling, computa-
tional modeling
RPI Image processing, SIMD Rich Sheroff
research
Harvard Machine Vision, Robotics Rich Sheroff
Univ of Bergen Dynamic modeling, DNA Neil Rowlands
sequencing, meteorology
CSIRO terrain surface display, Greg Madden/DEC
geophysics, mineral
exploration, meteorology
McGill object recognition, robotics Rich Sheroff
Univ of Bath numerical analysis Neil Rowlands
Texas Tech molecular dynamics, quantum Paul McManus
physics
Toyohashi Univ Extended common LISP Katsuei Aoyagi
implementation
Texas A&M Reservoir simulation, Paul McManus
molecular dynamics
Belcore Automatic indexing, document Rich Sheroff
retrieval, cryptography
continues on next page
MasPar installed sites continued
Customer Application MasPar sales rep
-------- ----------- ----------------
University of Akron Finite Element Analysis Greg Worth
Sumitomo Metals neural networks Katsuei Aoyagi
CMU Machine vision, robotics Greg Worth
Fregate/ENS sorting and searching tasks, Neil Rowlands
index table algorithms,
pattern matching
Johns Hopkins Univ Semiconductor modeling, Will Workman
radar signal processing
4. MPSG Notesfile and public directory
--------------------------------------
The DECmpp notesfile is now open at RDVAX::DECMPP.
We have also set up a public directory at
RDVAX::MPSG$PUBLIC. It contains the info sheet draft, the
MPSG newsletters, the DECmpp sales update article, the press
release, and a configuration guide. We will shortly copy the
overall DECmpp marketing presentation and the DECmpp product
presentation into the directory.
A1. DECmpp Sales Update Article with Prices and MPSG Contact List
-----------------------------------------------------------------
The DECmpp 12000 Massively Parallel Processing (MPP) Systems
CONTACT
Ram Appalaraju, DTN 223-4170, MLO
HIGHLIGHTS
o Among our competitors, Digital is the first major company to formally
enter the market for massively parallel computer systems
o Provides up to 16,384 processing elements capable of delivering 26,000
MIPS (32-bit add), 1.2 gigaflops 32-bit floating point
o DECstation 5000 front-end systems use industry standard ULTRIX operating
systems
o A robust software development environment through high-level language
data parallel programming
o Digital integrates MPP into a customer's distributed computing
environment
o Prices range from $264K -- $1.5M; delivery planned for September 1991
INTRODUCTION
Digital introduces a family of computers based on massively parallel
processing (MPP), an emerging, high performance technology targeted at the
very complex problems of technical, scientific and commercial users.
The new DECmpp 12000 system series features configurations ranging from
1,024 to 16,384 processors with performance up to 1.2 billions of floating
point operations per second (gigaflops), and 26,000 millions of instructions
per second (MIPS).
The new systems, supplied under a previously announced agreement with MasPar
Computer Corporation of Sunnyvale, California, are priced from $264,000 to
$1,502,500.
ONE CONSISTENT COMPUTING ENVIRONMENT
Digital now has a leadership position in high performance computing through
our DECmpp 12000 massively parallel processing systems, VAX 9000 systems,
networks and applications. We have the ability to integrate all of these in
one consistent computing environment for computationally intensive
applications.
Key to market acceptance, however, is software, ranging from application
development and network integration tools to the applications themselves,
that can take advantage of this technique.
Focus on Basic Software Elements
First, we're focusing on basic software elements such as language compilers,
libraries, networking designs and basic development tools. In some cases,
we're going to create parallelized versions of existing products; in other
cases, we're developing entire new products. For example, we are looking to
extend Digital tools such as our Language Sensitive Editor, Friendly UNIX,
and Test Manager to massively parallel architectures.
Stimulate Third-Party Applications
Secondly, we're committed to directly stimulating third-party application
development in strategic application areas.
Two application toolkits recently ported to these systems are "Elexir", an
information retrieval system from Third Eye Software, Inc. of Menlo Park,
California, and "Table-Maker", a statistical database package from CHISQ
Ltd., Reading, England.
Strategic Development Technologies
Digital is focusing its resources on strategic development technologies and
activities by:
o Developing and enhancing application development tools -- Digital will
extend development tool components of its CASE (Computer Aided Software
Engineering) product family to work with massively parallel
architectures; Digital will also pursue advance development in other key
technologies such as language compilers, data libraries and networking
technologies
o Stimulating third-party development activities -- Digital will pursue
partnerships and other initiatives to assist third-party application
developers with the creation of applications that take advantage of
massively parallel architectures
HOW MASSIVELY PARALLEL COMPUTERS WORK
Massively parallel processing is an outgrowth of the quest to speed computer
processing. The traditional computer architecture (known as the von Neumann
architecture) operates in serial fashion. With every cycle of a central
computer clock, the central processor unit (CPU) processes a fixed-size
instruction or word. The larger the word length, from 16 to 32 bits, for
instance, and the faster the processor, the greater the volume of
information that can be processed in a set time.
However, the faster the CPU, the more memory and larger input/output
bandwidth is required to maintain balanced, high-speed operations.
Multiple-processor implementations, with four, six, or even a hundred or
more processors, can bring significant increases in speed, but they must be
coordinated by additional logic in order to maintain synchronization. This
interprocessor overhead can add considerably to internal resource
utilization.
Massively parallel architectures employ thousands of relatively inexpensive
processor/memory units, distributed throughout the system. MPP
architectures are thus very efficient at handling large data arrays (of
1,000 by 1,000 elements, for instance), where the instruction flow can be
parallelized to operate on all elements simultaneously.
As a result, massively parallel computers are considered to have the
potential, as more application software becomes available, for handling a
broadening range of highly complex problem-solving applications.
PRODUCT DESCRIPTION
The DECmpp 12000 systems series consists of eight models, depending on the
number of Processor Element (PE) boards and memory per board. The DECmpp
12000 Series can be configured with 1024, 4096, 8192, or 16,384 processor
elements at initial installation or through on-site upgrades. By operating
in a Single Instruction Multiple Data (SIMD) fashion, the thousands of
processor elements can work on a single problem simultaneously. An entry
level DECmpp system provides 1,600 MIPS of processing power and is scalable
up to 26,000 MIPS -- and all within the same system package.
DECmpp Software Products
Although the DECmpp 12000 Series uses many interconnected processing
elements to achieve high performance, it features a single, integrated
software environment. This environment is based on the ULTRIX operating
system giving access to hundreds of utilities and applications. The user
interface to the software environment is based on the DECwindows system and
supports access from DECwindow terminals and DECstation workstations.
DECmpp Parallel Programming Environment
The DECmpp Parallel Programming Environment consists of a comprehensive set
of development tools for data parallel programming. This easy-to-use
environment features an interactive DECwindows-based point-and-select
interface. Users will easily visualize, analyze and optimize the behavior
of DECmpp processor elements, algorithms, and data. The DECmpp tool set
includes a symbolic debugger, a machine animator and data visualizer.
DECmpp Programming Language
DECmpp System software includes a C-like, high-level language compiler for
programming the DECmpp 12000 Series systems. The language is based on
Kernighan and Ritchie C and is the primary programming language for the
DECmpp system's data parallel unit.
DECmpp Image Processing Library
The DECmpp Image Processing Library is a highly optimized set of image
processing routines for the DECmpp 12000 Series Systems. The library
supports the development of very high performance image processing solutions
on a fully programmable platform with a symbolic, window-based, interactive,
graphical, and network-distributed development environment. Library
routines are callable from standard languages such as C, FORTRAN, and DECmpp
Programming Language. The library includes routines for image arithmetic,
low-level image processing operations, and image analysis, as well as a
variety of utilities for image display, I/O, test image generation, and
format conversion.
DECmpp Parallel Disk Array
The disk array provides the DECmpp 12000 Series systems with up to 11.2
gigabytes of formatted capacity and 9 MB/Sec sustained disk I/O. The disk
array is ideal for a wide range of data-intensive applications, such as
scientific visualization, remote sensing, image processing, and geographic
database management.
The disk array can be upgraded from four disks at 3 MB/Sec to 24 disks at 9
MB/Sec. Digital's TKZ08 tape drive may used to back up the parallel disk
array.
THE DIGITAL MPP STRATEGY
Digital has signed an OEM agreement with MasPar Computer Corporation to
resell MasPar's massively parallel computer systems, software and options.
The DECmpp 12000 systems combine excellent price/performance to provide a
highly productive growth path for the future. The product line is easily
scalable, upgradable and includes software tools for developing parallel
processing applications.
Digital's strategy is to:
o Provide an integrated environment by drawing on our strengths in areas
such as:
- Software development tools
- High performance networking
- System integration
- Workstations and servers
- Third-party software relationships
o Establish partnerships with leading MPP hardware and software companies
o Expand the market by raising awareness of the benefits of MPP
Why Did Digital Choose to Resell MasPar's Systems?
MasPar's products have high quality and an excellent price/performance. The
products are designed and built on state-of-the-art technology licensed from
Digital. This is the first partnership, but another, MIMD (Multiple
Instruction Multiple Data) partnership is planned for later this year.
PERFORMANCE POSITIONING
LINPACK
Peak 1K x 1K * Price
MIPS Number of MFLOPS 64-bit Range
32-bit Processors 64-bit MFLOPS $K
DECmpp 12000 26,000 16K 580 440 $ 274 -
Model 16B $1,533
Thinking 2,500 2K 14,000 5,200 $2,000 -
Machines CM-2 $7,000
Intel iPSC/860 1,056 32 1,300 486 $1,000 -
128 5,000 1,920 $4,000
VAX 9000 /410VP 30 1 125 79 $1,000 -
440VP 30 4 500 n/a $6,000
Cray Y-MP/132 N/A 1 333 324 $1,500 -
/832 8 2,667 2,144 $24,000
Convex C3210 N/A 1 50 17 $1,000
C3240 4 200 166 $6,000
C3810 1 106 99
* From report generated by Jack Dongarra, University of Tennessee, 6/12/91.
POSITIONING WITH RESPECT TO VAX 9000
The VAX 9000 vector processing system provides excellent performance in a
wide range of applications readily available from Independent Software
Vendors covering Science, Engineering, Chemical, Pharmaceutical and Oil and
Gas Industries. It also has excellent scalar and I/O performance, meaning
that it is a very balanced and versatile system to answer the growth needs
of the immense installed base of VAX system users in all segments. It will
very often be used as a compute or database server to satisfy the single
stream performance needs of an organization consisting of workstations or
time-sharing users on smaller VAX systems.
DECmpp 12000 Series system is a specialized execution engine capable of
extremely high performance for a more limited set of applications. Today,
it is relatively more difficult to program, but can yield dramatic execution
speed dividends on applications that involve computing on large arrays of
data. The DECmpp 12000 systems include a DECstation 5000/200, which
functions as a general purpose host computer to control and to execute
sequential code. VAX 9000 users can access the data parallel unit through
the DECstation workstation.
MARKETS AND TYPICAL APPLICATIONS
Markets for DECmpp systems include industries such as Government, Education,
Oil and Gas, Chemical, Pharmaceutical, Automotive/Aerospace, Service and
Financial, and Defense.
Typical applications include computational fluid dynamics, geophysical data
processing, structural dynamics and acoustics, ocean and atmospheric
modeling, computational chemistry and physics, simulation of neural nets,
molecular modeling, database and text retrieval, searching and sorting,
image processing and recognition, signal processing (1D, 2D, 3D), and
seismic processing and reservoir analysis.
The Massively Parallel Systems Group (MPSG) has an aggressive program to
provide off-the-shelf application software packages for DECmpp systems.
Please contact the MPSG staff for status and availability of application
software packages.
PRICING/ORDERING/CONFIGURATION/AVAILABILITY
All DECmpp 12000 Series systems include a DECstation 5000/200CX workstation
with an additional 16 MB RAM, VME adapter and SZ12B-FA sidecar, which
includes an RZ56 hard disk and a RRD42 CDROM drive, licenses for ULTRIX
workstation software and DECnet. Also included is a one-year Basic Service
warranty for all DECmpp 12000 Series systems.
The products listed below are also available from MasPar, but are priced
higher to represent added value when purchased from Digital. Digital's
prices include our comprehensive one-year warranty (versus MasPar's 90-day
return-to-factory warranty), our unlimited users per system software
licensing, our long-term commitment to support these products anywhere in
the world, and our investment in growing the value of these systems through
increasing application and system software.
RESOURCES
o Infosheets -- To order infosheets on DECmpp 12000 Series products,
contact Darlene Camuti at RDVAX::CAMUTI, DTN 223-3712.
o Marketing/Sales support staff -- for sales support for DECmpp
applications, configurations and competition for end user and volume
sales contact, one of the following DECmpp marketing staff:
Eastern U.S. Steve Harrington, DTN 223-3435 RDVAX::HARRINGTON
Western U.S. Eric Tyberg, DTN 521-6860, USWRLS::TYBERG_ER
Central U.S. Ron Renaud, RDVAX::RENAUD
Pan Europe Steve Zagame, 858-1067
United Kingdom Simon Cole, 899-3028
Germany, Austria,
Switzerland Jochen Krebs, 786-1408
France Yvan Bantoure, 858-1397
South Pacific, Asia Greg Madden @SNO, DTN 730-5454, (61-2-561-5454)
European product mgr Bob Pattison, 899-5734
Japan Hiroshi Mizoguchi, 03-989-7340
o Service contacts Diana McClure, 264-0883, MVDS02::MCCLURE
Mike Carlton, 264-5740, MVDS02::CARLTON
o Additional Contacts -- Ram Appalaraju, Program Manager,
RDVAX::APPALARAJU, DTN 223-4170; Bob Nilsson, MPSG Sales & Marketing
Manager, RDVAX::NILSSON, DTN 223-1715; Charlie Wilson, MPSG Group
Manager, DTN 223-2867.
o Order processing contact -- DECmpp products must be quoted on AQS prior
to order entry and the quotes sent to Lisa Peloquin at RDVAX::PELOQUIN.
For help with quoting and processing orders, contact Lisa Peloquin, DTN
223-1935, RDVAX::PELOQUIN.
o All the DECmpp products have a lead time of 90 days for domestic orders
and 120 days for international orders.
DISCOUNTS
All DECmpp products are discountable under the DBA, category A (0 - 21%) for
both end users and resellers. Also, the products can be sold under GSA
contracts.
SERVICE
All DECmpp 12000 Series systems, software and options are fully supported by
Digital Services organizations as standard Digital products. Due to the
specialized nature of DECmpp 12000 Series systems, all customer and Volume
sales opportunities should be qualified through the MPSG technical staff.
Contact Lisa Peloquin @MLO DTN 223-1935 to qualify and configure customer
orders.
DECmpp 12000 Series Systems, Available September 1991
Model No. Description U.S. List Std.
MP101-AA Model 1A, 1K PE's, 16MB RAM, 120V $ 240,500 $ 264,300
MP101-AB Model 1A, 1K PE's, 16MB RAM, 240V 240,500 264,300
MP101-AC Model 1A, 1K PE's, 16MB RAM, 240 SH 240,500 264,300
MP102-AA Model 1B, 1K PE's, 64MB RAM, 120V $ 275,000 $ 299,400
MP102-AB Model 1B, 1K PE's, 64MB RAM, 240V 275,000 299,400
MP102-AC Model 1B, 1K PE's, 64MB RAM, 240 SH 275,000 299,400
MP103-AA Model 4A, 4K PE's, 64MB RAM, 120V $ 378,500 $ 413,000
MP103-AB Model 4A, 4K PE's, 64MB RAM,240V 378,500 413,000
MP103-AC Model 4A, 4K PE's, 64MB RAM, 240 SH 378,500 413,000
MP104-AA Model 4B, 4K PE's, 256MB RAM, 120V $ 516,500 $ 558,500
MP104-AB Model 4B, 4K PE's, 256MB RAM, 240V 516,500 558,500
MP104-AC Model 4B, 4K PE's, 256MB RAM, 240SH 516,500 558,500
MP105-AA Model 8A, 8K PE's, 128MB RAM, 120V $ 540,000 $ 589,000
MP105-AB Model 8A, 8K PE's, 128MB RAM, 240V 540,000 589,000
MP105-AC Model 8A, 8K PE's, 128MB RAM, 240SH 540,000 589,000
MP106-AA Model 8B, 8K PE's, 512MB RAM, 120V $ 804,000 $ 860,000
MP106-AB Model 8B, 8K PE's, 512MB RAM, 240V 804,000 860,000
MP106-AC Model 8B, 8K PE's, 512MB RAM, 240SH 804,000 860,000
MP107-AA Model 16A, 16K PE's, 256MB RAM, 120V $ 892,000 $ 970,000
MP107-AB Model 16A, 16K PE's, 256MB RAM, 240V 892,000 970,000
MP107-AC Model 16A, 16K PE's, 256MB RAM, 240SH 892,000 970,000
MP108-AA Model 16B, 16K PE's, 1GB RAM, 120V $1,420,000 $1,533,000
MP108-AB Model 16B, 16K PE's, 1GB RAM, 240V 1,420,000 1,533,000
MP108-AC Model 16B, 16K PE's, 1GB RAM, 240SH 1,420,000 1,533,000
NOTE: 1 The data parallel unit in all the models requires 240V 50/60 HZ
power source. The voltages indicated in the above description are
for the workstation only.
2 MP10X-AC models include monitors for use in the Southern
Hemisphere.
DECmpp PE Array Options, Available September 1991
KP100-AA 1K PE board with 16 MB RAM $ 51,750 $ 55,500
KP100-AB 1K PE board with 64 MB RAM 86,250 92,300
KP100-AC 1K PE board upgrade from 16MB to 64MB 86,250 90,100
DECmpp Interface Channel Option, Available September 1991
KF100-AA DECmpp I/O Controller Option W/8MB $ 21,600 $ 23,660
DECmpp Parallel Disk Array Systems, Available September 1991
SMP01-AA Parallel disk array system w/4 data $ 129,800 $ 145,000
disks, 2.9GB. requires KF100-AA
SMP01-AB Parallel disk array system w/8 data 176,000 192,000
disks, 5.8GB. requires KF100-AA
SMP01-AC Parallel disk array system w/16 data 237,600 253,000
disks, 11.5GB. requires KF100-AA
DECmpp Disk Array Upgrade Options, Available September 1991
SMP02-AA SMP01-AA to SMP01-AB upgrade $ 51,700 $ 55,200
SMP02-AB SMP01-AA to SMP01-AC upgrade 118,800 127,200
SMP02-AD SMP01-AB to SMP01-AC upgrade 67,100 72,100
DECmpp Hard Disk, Available September 1991.
RMPO1-AA Hot Standby Disk and Controller $11,550 $12,900
DECmpp Software Products, Available September 1991
Standard (Capacity
Lic. Per System)
QL-XT4A8-AA DECmpp System Software License $ 120
QA-XT4AA-GZ Documentation for DECmpp System Software (Available 9/25)
QL-XT7A8-AA DECmpp Parallel Programming $4,710
Environment License
QA-XT7AA-GZ Documentation for DECmpp Parallel (Available 9/25)
Programming Environment software
QL-XT5A8-AA DECmpp Image Processing Library License $4,710
QA-XT5AA-GZ Documentation for DECmpp Image (Available 9/25)
Processing Software Library
QA-XT4AA-H8 CDROM Media for DECmpp System Software $ 335
DECmpp Parallel Programming Environment,
DECmpp Image Processing Library
QA-XT4AA-H5 TK50 Media for DECmpp System Software $ 225
DECmpp Parallel Programming
Environment, DECmpp Image Processing
Library.
A2. The DECmpp Press Release
-----------------------------
Joseph D. Codispoti
(508) 493-6767
DIGITAL EQUIPMENT INTRODUCES FAMILY OF
MASSIVELY PARALLEL COMPUTERS
MAYNARD, Mass. -- August 20, 1991 -- Digital Equipment Corporation
today introduced a family of computers based on massively parallel
processing (MPP), an emerging, high performance technology targeted
at very complex problems of technical, scientific and commercial
users.
The new DECmpp 12000 system series features configurations
ranging from 1,024 to 16,384 processors with performance up to 1.2
billions of floating point operations per second (GigaFLOPS), and
26,000 millions of instructions per second (MIPS).
The new systems, supplied under a previously announced
agreement with MasPar Computer Corporation of Sunnyvale, California
are priced from $240,500 to $1,502,500. Worldwide shipments will
begin in the fall.
All DECmpp systems are configured with Digital DECstation 5000
full color front-end workstations, as well as ULTRIX and DECnet
software licenses. In addition to the family of computer
processors, Digital also announced a comprehensive line of
software, a family of high speed disk arrays, and full service and
support programs. Digital's standard one year product warranty
will apply.
With this announcement, Digital becomes the first main line
systems vendor to formally introduce a family of massively parallel
systems, and it is the first time users of systems in this
performance class can take advantage of Digital's high level of
warranty, service and support.
Charles Wilson, director of Digital's Massively Parallel
Systems Group, noted "We feel this is a milestone in the evolution
of massively parallel technology. There are many businesses,
government agencies and universities that want to move to massively
parallel technology, but are looking to more application software
and higher levels of enterprise service and support.
"Digital's entry in the market is important to these users for
two reasons. First, we will stimulate application program
development both directly through our independent software
developers, and indirectly by adding the commitment of an
established, major systems vendor to this relatively new market.
"The second reason is Digital's leadership in delivering
effective, highly reliable, enterprise-wide support to large
organizations. MPP systems are yet another important technology to
be integrated into the total computing environment."
Introduced today were:
o DECmpp 12000 systems -- eight field-upgradable models
incorporating 1,024, 4,096, 8,192 and 16,384 processor
elements, and memory capacity from 16MB to 1GB. Priced from
$240,500 to $1,502,500;
o DECmpp Programming Language -- a C-like high level compiler
for programming DECmpp systems included with the DECmpp
System Software license, $100;
o DECmpp Parallel Programming Environment -- a comprehensive
suite of development tools for data-parallel programming.
Features an interactive DECwindows point-and-select
interface. Includes symbolic debugger, machine animator and
data visualizer. License, $4,000;
o DECmpp Image Processing Library -- a facility for supplying
common visual and data objects to help speed program
development. License, $4,000.
o DECmpp FORTRAN -- built on FORTRAN 90, this compiler will
provide facilities to program in high level language and
generate code for data parallel execution. Available in
December;
o DECmpp Disk Array Systems -- high speed disk-array
subsystems, provide parallel access to large data files.
The disk array features 720 megabytes of formatted storage
per disk with a capacity up to 11.2 gigabytes of storage
and sustained I/O rate of nine megabytes per second.
Priced from $129,000 to $261,000;
o Digital Service -- the complete worldwide service offering
from Digital, including enterprise integration, consulting,
customer training, and maintenance and repair services, is
available to DECmpp customers.
SOFTWARE AIDS MARKET GROWTH
Massively parallel processor architectures are considered by
many industry experts to have the greatest potential for solving
highly complex problems in technical, scientific and commercial
computer applications such as scientific visualization, geographic
database management, econometric modeling, and financial analysis.
MPP computers are able to break down problems into smaller
parts, then process these parts simultaneously, at very high speed.
As a result, performance can be extremely fast for applications
that lend themselves to this type of processing.
Key to market acceptance, however, is software, ranging from
application development and network integration tools to the
applications themselves that can take advantage of this technique.
Digital is working to advance MPP software development on
several fronts, according to Wilson.
"First, we're focusing on basic software elements such as
language compilers, libraries, networking designs and basic
development tools. In some cases, we're going to create
parallelized versions of existing products; in other cases, we're
developing entire new products. For example, we are looking to
extend Digital tools such as our Language Sensitive Editor,
Friendly UNIX, and Test Manager to massively parallel
architectures.
"Secondly, we're committed to directly stimulating third-party
application development in strategic application areas."
Targeted applications include: structural dynamics in
automotive and aerospace industries; seismic processing and
reservoir analysis in oil and gas exploration; and database
retrieval, searching and sorting.
Two application toolkits recently ported to these systems are
"Elexir", an information retrieval system from Third Eye Software,
Inc. of Menlo Park, California, and "Table-Maker", a statistical
database package from CHISQ Ltd., Reading, England.
DECmpp EXTENDS HIGH PERFORMANCE SOLUTIONS
The new DECmpp 12000 Series Systems add an important MPP
component to Digital's high-performance computing products.
Digital currently manufactures and markets VAX 9000 and VAX 6000
systems, including mainframes and vector equipped versions for
supercomputing. Digital also offers a range of high-performance
workstations. These and other Digital products maintain industry
leading interoperability with non-Digital systems via the company's
open NAS architecture of standard software interfaces.
Digital Equipment Corporation, headquartered in Maynard,
Massachusetts, is the leading worldwide supplier of networked
computer systems, software and services. Digital pioneered and
leads the industry in interactive, distributed and multi-vendor
computing. Digital and its partners deliver the power to use the
best integrated solutions -- from desktop to data center -- in open
information environments.
####
CORP/92/284
Note to Editors: MasPar is a trademark of MasPar Computer
Corporation.
Table-Maker is a trademark of CHISQ Ltd.
Elexir is a registered trademark of Third Eye
Software, Inc.
UNIX is a trademark of Unix System Laboratories,
Inc.
DECnet, DECstation, DECmpp, VAX and ULTRIX are
trademarks of Digital Equipment Corporation.
CURRENT APPLICATIONS
The newer massively parallel architectures make it possible to
process computationally intensive applications with ultra high
speed and relatively low cost. For example, a 16,000-processor
machine is rated at 26,000 peak MIPS and 1,200 peak MFLOPS, and
costs just over $1 million.
Some typical applications are:
o Structural analysis -- used frequently in automotive and
aerospace industries, this relates to finite element
analysis of mechanical structures, both at rest and under
stress;
o Computational fluid dynamics -- the fastest growing area of
computational physics, this typically involves the study of
aerodynamics, including airflow over airfoils and air
resistance of automobiles;
o Seismic/reservoir modeling -- these are the two most used
oil-industry applications;
o Signal and image processing -- used extensively by NASA and
the military, this application requires very fast filtering
and analysis of electronic signals;
o Chemistry/biology -- these applications include studies of
chemical structures, reactions and properties, as well as
genetic molecular modeling;
o Commercial -- the major potential today for business/office
users involves database, financial-modeling and text-
retrieval applications.
|
|
D I G I T A L
- ____ ____ ____ ____
- / / / /___/ /____ / __
- / / / / ____/ /____/
Computers of the future...
...available today
MPSG NEWSLETTER #6
Massively Parallel Systems Group - December Update
***Digital Internal Use Only***
Attached is the sixth update from MPSG on Digital's MPP program and
plans. Please let Darlene Camuti (RDVAX::CAMUTI) know if you would
like to be deleted from this distribution list or if you know of
someone who should be added.
Contents
--------
1. Program Summary
2. Product Update
3. Training Plans
4. Applications
5. MPP Positioning Review
6. DECmpp Contacts
7. Recent and Upcoming DECmpp Events
8. Press Release: Georgia Tech DECmpp sale
9. Press Release: New MPSG Staff Members
10. Sales Update Draft on DECmpp FORTRAN
11. How To Access DECmpp User Documentation
1. Program Summary
---------------
DECmpp is Digital's entry into the massively parallel
processing market. The DECmpp 12000 is based on a
single-instruction, multiple data (SIMD) architecture. The
products, prices and availability were announced on August
20, 1991. These products are currently on Digital's
corporate price list worldwide.
MPP systems offer both the maximum computing performance
available and the maximum performance at a given price
level for applications involving large arrays of data.
Examples of these applications include database searching
and sorting, computational chemistry, automobile design,
statistical analysis, text retrieval, signal and image
processing, financial analysis, and seismic processing.
Digital's strategy is sell midrange MPP systems which
integrate into the customers' computing environment. This
integration is accomplished by extending NAS and the
Cohesion toolset to embrace our MPP systems. Digital's major
commitment to this software environment is evidenced by
MPSG's nineteen software engineers developing and enhancing
our MPP compilers and toolsets. We have also assembled a
team of application consultants to port and optimize
applications for DECmpp.
Our DECmpp platforms range from 1,024 to 16,384 processors
per system and are priced from $250,000 to $1.5M. Peak
performance of the large system is 1.3 Gigaflops.
Selling DECmpp
Target customers typically have 1-3 large, time consuming
computer applications which require hours or days to execute
on traditional systems. MPSG is prepared to accompany you on
visits to these customers to present the benefits of MPP
and show how our software integrates via Network
Application Support. If the customer's application is not
yet available off-the-shelf, we have experts who can
determine how easily the customer's application can be
ported to DECmpp.
Note that while the DECmpp SIMD hardware was designed by
MasPar Computer Corporation, please do not call MasPar
sales reps to help with your sales. The Digital and MasPar
software environments are diverging and the MasPar reps will
not be able to accurately represent the Digital products.
The sales support staff at MPSG has ramped up to 25 people
worldwide, including PhDs with extensive MPP experience. We
guarantee quick response to telephone or electronic
requests.
DECmpp Sales Tools
------------------
o DECmpp info sheet (EC-F1545-15)
These info sheets are orderable through Northboro
Publishing and Circulation Services.
o Digital High Performance Computing brochure (EC-E1145-57)
o DECmpp customer presentation
(RDVAX::MPSG$PUBLIC:customer_presentation.ps)
o DECmpp configuration guide
(RDVAX::MPSG$PUBLIC:decmpp_configuration_guide.txt)
o Notesfile - RDVAX::DECMPP
o Hotline mail account - RDVAX::DECMPP
o Product manuals (all docs are now complete!) -
in the public directory: RDVAX::MPSG$PUBLIC:decmpp_docs.dir
o Demos - available on request, software available on
RDVAX::MPSG$PUBLIC:decmpp_demos.tar
2. Product Update
--------------
DECmpp Fortran (Ql-XT6A8-AA) is scheduled for announcement
at the end of this month. The target price is $10,000. The
field test version of this software is currently being
shipped to customers. The FRS will be February 15, 1992.
See attached Sales Update draft for more info on DECmpp
FORTRAN.
These documentation kits have been added to the price list
as of 12/2/91:
QA-XT4AA-GZ DECmpp SYSTEM SOFTWARE DOCUMENTATION $465
QA-XT5AA-GZ DECmpp IMAGE PROCESSING LIBRARY $136
DOCUMENTATION
QA-XT7AA-GZ DECmpp PARALLEL PROGRAMMING ENVIRONMENT $643
DOCUMENTATION
Product enhancements planned over the next 4-6 months:
o Announcement of VAST II for optimizing MPP software
o Repackaging DECmpp 12000 into new Digital standard
enclosure system
o Implementation of DECmpp software environment based on
FUSE and COHESION
Product enhancements planned over the next 6-15 months:
o Introduction of new High Performance/Data Parallel Fortran Compiler
o Introduction of MIMD system to complement SIMD offering
o Boost in performance of current SIMD system 4-10 times
3. Training
--------
The next class of DECmpp sales and support training is now
scheduled for January 8-10 in Maynard, MA. Please send a
note to Darlene Camuti to register for the class. Student
reviews of this training have been excellent and we have had
to limit attendance at each class given to date, so sign up
early. The training agenda for this class is below.
DECmpp Product Training
January 8 - 10 1991
Dan Cook Conference Room at PK3-2/26c
DAY 1: Jan 8
8:30 Introduction to MPSG and training goals
9:00 MPP style of computing, Digital strategy and plans
10:30 break
10:45 MasPar background
11:00 DECmpp Products
11:30 lunch
12:30 Demo part 1 - sorting, Mandlebrot
1:00 MPP product training part 1
2:30 Break
2:45 MPP product training part 2
4:15 MPP positioning and Day 1 wrap up
4:30 Demo part 2 (MRI, MPPE)
5:00 end of day 1
DAY 2: Jan 9
8:30 MPP application software
9:00 Selling MPP systems vs the competition
10:00 break
10:30 Marketing events
10:45 Configuring, quoting, site prep, Digital MPP mfg
11:15 Customer support
12:00 Lunch
1:00 Hardware futures
1:30 Software futures
2:00 break
2:15 discussion and wrap up
DAY 3: Jan 10
1x1 Consulting on customer opportunities.
Optional MPP programming instruction.
Quiz.
DECmpp field seminars
---------------------
If you would like to hold a DECmpp seminar for your customers,
we would be glad to help out. Here are three example seminar
topics: "DECmpp and Large Scale Statistical Analysis",
"DECmpp and Automotive Crash Test Simulation", and
"Digital's Massively Parallel Processing Strategy and
Products".
We ask only that you assign someone from your sales office
to be responsible for the logistics and customer invitations
and that you help pull together a training session for your
local support team. We will supply a presentor and demo. If
you would like to pursue this please call or send mail to
the MPSG marketing rep assigned to your geography (see
section 6 of this newsletter).
Two new training courses are being offered by Digital
Educational Services. Additional dates will be scheduled as
requested.
SERVICES TRAINING
NAME: DECmpp SOFTWARE TRAINING (VENDOR TRAINING)
COURSE CODE: EY-H997E-S0
DATES: DECEMBER 2-5, 1991
DURATION: 4 DAYS
FORMAT: LECTURE/LAB
LIMIT: 10
TUITION: $1,000.00
START TIME: 9:00 A.M.
NAME: DECmpp SYSTEMS MAINTENANCE and DECmpp DISK ARRAY
MAINTENANCE TRAINING (VENDOR TRAINING)
COURSE CODE: EY-H998E-S0
DATES: DECEMBER 9-13, 1991
DURATION: 5 DAYS
FORMAT: LECTURE/LAB
LIMIT: 6
TUITION: $2,082.50
START TIME: 8:30 A.M. (Please note earlier start time.)
DESCRIPTION INFORMATION - DECmpp SOFTWARE TRAINING
Prerequisites: Some familiarity with SIMD computer architectures C
Programming with intensive coverage of syntactic and semantic extensions to
C, MPPE and System Administration.
Topics Covered:
. Array Control (ACU) concepts
. ACU-to-Procesor and Processor-to-ACU communications
. Data paths between Front End workstation and ACU Processors
. Processor on-ship and off-ship memory concepts
. XNET (nearest neighbor) inter-processor communication
. Global Router inter-processor communication
. High-speed I/O data paths
. Plural data paths
. mp_fort, mpi_cc and stand along mpl programming models
. Compile-time options; DPU/workstation compatibility issues
. Plural aggregates, arrays, and pointers
. Plural loop, conditional, switch, and procedure returns
. Interaction between plural/singular loops, conditionals, etc.
. Processor selection and inhibition
. Pseudo-structures proc[], xnet[], and router[]
. Library routines for parallel arithmetic functions
. Library routines for special parallel constructs
. Library routines for workstation/DPU communication and control
. Brief overview of library routines for high-speed I/O
. Window, cursor, and mouse button conventions in MPPE
. Principal debugger features and menu options
. Remote procedure calls, X Windows - Yellow Pages interactions
. Hands-on exercises: walk through of MPPE tutorial
. directory trees
. device driver/config overview
. dpumanager execution and log files
. networking/yellowpages issues
. software installation procedure
DESCRIPTION INFORMATION - DECmpp SYSTEMS MAINTENANCE and DECmpp DISK ARRAY
MAINTENANCE TRAINING
Please Note: This training combines a 4-day and a 3-day course into one
5-day intensive training.
Prerequisites: Experience in hardware troubleshooting techniques and ULTRIX
system administration.
Topics Covered:
. System Architectural Overview
. Theory of Operation
. System installation and verification
. Performing important ULTRIX system administration functions
. Installation of optional products
. System diagnostics
. Field replaceable unit replacement procedures
. Performing electronic adjustments
. Documentation review
. Theory of operation
. System installation and verification
. Performing important ULTRIX system administration functions
. System diagnostics
. Field replaceable unit replacement procedures
. Performing electronic adjustments
. Performing disk array upgrades to larger models
---------------------------------------------------------------------------
LOCATION: MasPar Computer Corporation, 749 North Mary Avenue, Sunnyvale, CA
ENROLLMENT INFORMATION: Call CENTRAL
REGISTRATION, DTN 264-7000 or (603) 884-7000. Direction information will be
included in the employee's confirmation letter.
TO WRITE: Today only, write to BUOREG @BUO or ESMAIL::BUOREG, after today,
write to: BUOREG @DDD or CGVAX2::BUOREG.
Include: Your name, badge, CC, DTN, Mail Node, Mgr's Name and Mail Node,
Course Name, EY Course Code, and Course Date.
4. Applications
------------
The following are off-the-shelf codes available for DECmpp:
SAS/TM ACCESS (CHISQ Ltd)
This software provides the large-scale statistical analysis
capabilities of Table-maker integrated into the popular SAS
statistical software package. It provides very high speed
statistical database functions such as data correlation,
averaging and histogramming. The package effectively
replaces mainframes from IBM, ICL, CDC which are currently
being used for statistical analysis. The developer, Ian
Taylor (reachable at dtn 849-3614) located in Reading,
England, is available to provide presales support and
customer consulting for this application.
ELEXIR (Third Eye Software, Inc.)
Elexir is a toolkit for building information retrieval
systems which achieve unprecedented performance for
similarity searches of unstructured text in applications
including legal and financial services, wire information
services, message routing and customer support. Elexir on
the DECmpp 12000 is capable of searching 8 million documents
per second.
Radioss (Mecalog)
Radioss is a car crash simulation code which uses the finite
element method. It achieves 1/3 the performance of a YMP
processor on the DECmpp 4K processor system. This code will
be available for customers starting in January.
Codes in process
----------------
The following is a sample of the most promising codes to be
ported to DECmpp systems over the next 3-9 months. We'd be
glad to work with customers who are interested in using
DECmpp systems to run these codes.
CHARMm (Molecular Simulations, Inc)
The CHARMm software package models the static and dynamic
behavior of molecular systems. The code uses empirical
energy functions to compute energies, find minimum-energy
conformations, determine the energy cost of deformations and
simulate vibrational motions. The code is applicable to
small organic and drug molecules, peptides and proteins,
nucleic acids and other biopolymers, and a variety
of synthetic polymers and inorganic materials.
FIDAP (Fluid Dynamics International, Inc)
FIDAP uses the finite-element method to simulate classes of
incompressible fluid flows. The code provides a range of
possible analyses, including isothermal and non-isothermal,
Newtonian and non-Newtonian flows; turbulent flows; free,
forced or mixed convection flows; periodic; separating or
recirculating flows; swirling flows; creeping flows; flows
in rotating frames of reference; flows with a free or moving
surface; and surface tension, gradient-driven thermal flows.
The following codes are under investigation (for short term
conversion, if feasibility is verified).
o Amber
o Xplor
o AVL/FIRE
o Mathcad
o IMSL
o DYNA3D
o AVS
The following are longer term code conversion efforts. These
are high priorities, but, if feasible, will take more than a
year to port. We are currently negotiating business
arrangements with the code providers to port and support the
codes. This is for your information only; we need to
progress further with these before this information can be
shared with customers.
o ABACUS by HKSI
o Gaussian 9X
5. Product Positioning
-------------------
With Alpha systems coming and negotiations to sell the Cray
YMP-EL progressing, it is important to understand how to
position DECmpp systems among RISC and vector products.
DECmpp systems offer the highest performance for a given
price level for applications that map well to MPP. Current
performance is up to 1.3 GF. The next DECmpp SIMD
generation, to be available next year, increases this 4-10
times. Significantly, DECmpp software is a full member of
Digital's NAS and Cohesion product lines. Its operating
system is ULTRIX, it uses DECwindows, the CASE environment
will be moving to DEC FUSE and all networking is based on
standard Digital products. The DECmpp customers' growth path
is within the Digital product line which will grow to also
include MIMD MPP.
Applications which map well to MPP include:
Text retrieval financial modeling
large-scale statistical analysis computational fluid dynamics
structural analysis algorithm research
computational chemistry seismic analysis
image and signal processing reservoir analysis
RISC systems (including Alpha in the future) offer medium
high performance over the widest range of applications,
especially those which are less data-intensive. RISC offers
higher performance than MPP for applications which are less
data intensive such as: software development, electronic
mail, office automation, personal productivity, MRP and
numerically-intensive applications that do not involve
arrays of data. Performance of high end Alpha systems (based
on EV-4 to be announced next calendar year) will be in the
range of approximately 70-150 MFLOPS per processor.
VECTORS offer an attractive compromise between highest
speed (of MPP) and the widest range of applications (of
RISC). Many scientific applications have already been
optimized for vector processors. Data-intensive applications
which do not fit well into the MPP application categories
listed above may fit vector processors such as the VAX
6000VP and VAX 9000VP.
The YMP-EL will be a good fit for customers who require full
Cray YMP compatibility and need to run a range of highly
vectorized UNIX applications. Maximum performance of the
YMP-EL is 133 MFLOPS per CPU, up to four CPUs. Selling
YMP-ELs allows us to retain account control where the
customer is committed to purchasing a fully CRAY compatible
mid-range system. Note that while negotiations for Digital
to resell YMP-EL systems are progressing rapidly, an
agreement has not yet been signed.
6. DECmpp Contacts
---------------
Eastern U.S. Steve Harrington, 223-3435, RDVAX::HARRINGTON
Western U.S. Eric Tyberg, 521-6860, USWRLS::TYBERG_ER
Central U.S. Ron Renaud, 223-4161, RDVAX::RENAUD
Pan Europe Steve Zagame, 858-1067, CCIIS1::ZAGAME
United Kingdom Simon Cole, 899-3028
Germany, Austria,
Switzerland Jochen Krebs, 786-1408, MTS$::"unt::jochen KREBS"
France Yvan Bantoure, 858-1397,MTS$::"FGT::YVAN Bantoure"
Nordic countries Magnus Persson, 876-8000, mts$::"soo::magnus persson"
South Pacific, Asia Greg Madden @SNO, DTN 730-5454, (61-2-561-5454)
European product mgr Bob Pattison, 899-5734, CHEFS::PATTISONB
Japan Hiroshi Mizoguchi, 03-989-7340, TKOVOA::mizoguchi_h
o Service contacts Diana McClure, 264-0883, MVDS02::MCCLURE
Mike Carlton, 264-5740, MVDS02::CARLTON
o Additional Contacts --
Ram Appalaraju, Program Manager, RDVAX::APPALARAJU, 223-4170
Bob Nilsson, MPSG Sales & Mktg Manager, RDVAX::NILSSON, 223-1715
Charlie Wilson, MPSG Group Manager, RDVAX::WILSON, 223-2867
DECmpp Application Consultants
Bill Celmaster, 223-2173, RDVAX::CELMASTER
Ajit Agrawal, 223-2383, RDVAX::AGRAWAL
George Anagnostou, rdvax::anagnostou
Bob Monroe, rdvax::monroe
o Order processing contact -- DECmpp products are available
for quoting in the AQS system. For help with quoting and
processing orders, contact Lisa Peloquin, DTN 223-1935,
RDVAX::PELOQUIN.
o In-field support people who have received DECmpp training
Over sixty in-field sales support people have been trained
on DECmpp. Give the MPSG support person assigned to your
geography a call for help in assembling a team for your
sales opportunity.
7. DECmpp events
-------------
November 16-18 Japan DECmpp training Tokyo
November 10-14 DECmpp at SEG show, Houston
November 14-16 DECmpp at IEEE Supercomputing '91, Albuquerque
November 26 DECmpp presentation MRO4 Amphitheater
December 3 Atlanta, Georgia DECmpp training
December 5 Novi, Michigan DECmpp training
January 8-10 DECmpp sales and support training
February 12-14 Supercomputing Europe, Paris
mid-April Supercomputing Japan
Apr 27-May 15 DECworld - including DECmpp sessions and demos
8. Press Release: Georgia Tech DECmpp Sale
Joseph D. Codispoti
508-493-6767
Molly Croft
Georgia Institute of Technology
404-853-2682
GEORGIA TECH GETS FIRST DIGITAL EQUIPMENT MASSIVELY PARALLEL
SYSTEM TO FURTHER RESEARCH INTO THIS EMERGING ARCHITECTURE
ATLANTA, GA -- November 19, 1991 -- The first installed
massively parallel computer from Digital Equipment
Corporation is being used by researchers at Georgia Institute
of Technology to better understand and advance the use of
this emerging computing architecture.
The DECmpp 12000, installed at the recently formed
College of Computing, joins other advanced parallel computers
as part of research into system dynamics related to software
design, input/output subsystems, data management, storage,
and interprocessor communication as computers scale upwards
from a handful of processors to several thousand.
Research in the College, thus far, has been focused on
small to medium scale parallel computers with fewer than
1,000 processors where a critical issue is internal
communication among processors.
According to Dr. Kishore Ramachandran, associate
professor in the College, "The DECmpp 12000 can scale up to
16,000 processors allowing us to focus attention on
input/output and other architectural issues related to such
large scale parallelism.
"The DECmpp 12000 is our first SIMD style machine. The
programming environment of the DECmpp 12000 makes it
extremely easy to introduce parallel computing to our
students." He added that the DECmpp 12000 was also an
affordable parallel machine, that was easily installed in
August and ready for September classes.
Dr. Ramachandran, a 1990 NSF Presidential Young
Investigator, said the College has been engaged in
considerable research into distributed computing
architectures and is now aggressively looking at massively
parallel structures. In addition to specific research in
this area, the DECmpp system is also being used for imaging,
computer vision and numerical algorithms.
Dr. Peter Freeman, Dean of the College of Computing
said, "The acquisition of the DECmpp 12000 is part our
ongoing effort to establish a high performance computing
experimentation laboratory in the College for the entire
campus. It was funded through this College and the School of
Electrical Engineering as well as a grant from Digital. The
mission of the laboratory is to develop high performance
computing software and hardware by collaborating with
researchers from a number of application fields."
The DECmpp 12000 at Georgia Tech features 1024 4-bit
processors which can be expanded to more than 16,000. Fully
expanded, the system can perform up to 1.2 billions of
floating point operations per second (GigaFLOPS) and 26,000
millions of instructions per second (MIPS). Price for the
DECmpp 12000, which was introduced in August, range from
about $240,000 to $1.5 million.
Digital Equipment Corporation, headquartered in Maynard,
Massachusetts, is the leading worldwide supplier of networked
computer systems and services. Digital offers a full range
of computing solutions and systems integration for the entire
enterprise -- from the desktop to the data center.
####
10. Press Release: New MPSG staff members
Joseph D. Codispoti
508-493-6767 11/10/91 Sun 0:34:29
DTN-223-6767 11-Nov-1991 Mon 12:33
DIGITAL EQUIPMENT ASSEMBLES TOP TECHNICAL TEAM
TO IMPLEMENT ITS MASSIVELY PARALLEL STRATEGY
MAYNARD, MA -- November 19, 1991 -- Digital Equipment
Corporation has recently named top technical professionals to
lead the implementation of its strategy for leadership for
massively parallel computing systems. Digital announced its
entry into this emerging market in April, and introduced its
DECmpp 12000 systems series in August with the expressed strategy
of accelerating the development and use of software to help
stimulate the market.
New senior staff include: Ching-Ching Ganley, Engineering
Manager; Marco Annaratone, Technical Director; William Celmaster,
Applications Consultant; Ajit Agrawal, Applications Consultant.
'We have assembled a group of extraordinary individuals who
will each contribute significantly to our software strategy which
is the key to the broad acceptance of massively parallel
technology", said Charles Wilson, Manager of Digital's Massively
Parallel Systems Group. "They join other top professionals
already in place and they will work collaboratively with the
extensive and talented resources within Digital's worldwide
research and engineering organization."
Ching-Ching Ganley has been with Digital for nearly 20
years having increasing responsibilities in software engineering
and engineering management with TOPS 10, display systems
software and networking software including OSI and distributed
systems. In her most recent assignment she was responsible for
strategic technology and planning functions for the Publishing
and Distribution Technology Group. Prior to joining Digital, she
had been with Honeywell Information Systems in software
engineering, development and consulting. Ching-Ching has a BS in
mathematics from Pu Jen University {CITY, COUNTRY} an MS degree
in mathematics from Clarkson College and MBA from Northeastern
University.
Marco Annaratone joins Digital from the Integrated Systems
Laboratory, Swiss Federal Institute of Technology in Zurich,
Switzerland, where he has been an assistant professor of computer
science. Dr. Annaratone is a highly regarded computer scientist
having conducted research in areas of VLSI architectures,
massively parallel architectures, and, earlier, in the areas of
speech compression and fault tolerance. Widely published in
technical journals and textbooks, he had previously been a
researcher at Carnegie Mellon University and Politecnico di
Milano where he earned the Dott. Ing. degree.
William (Bill) Celmaster comes to Digital from BBN/Advanced
Computers, Inc. where he was Chief Scientist and North American
Analyst Manager. Prior to BBN, Celmaster had been a consultant
with Symbolics, Inc. and an assistant professor of physics at
Northeastern University and University of Illinois. His areas of
expertise include parallel and vector programming, mathematical
and scientific algorithms and fluency in 'C' and FORTRAN. He
holds a BS in mathematics and physics, and earned the MS and PhD
degrees in physics at Harvard University.
Ajit Agrawal had been with Thinking Machines Corporation for
several years before joining Digital. While there, he designed
and implemented a parallel fault simulator for testing VLSI
circuits for the Connection Machine and also designed many of the
key parallel primitives for this machine. He earned his
undergraduate engineering degree at the Indian Institute of
Technology, and the MS, Masters of Philosophy and PhD in Computer
Science at Brown University.
Digital Equipment Corporation, headquartered in Maynard,
Massachusetts, is the leading worldwide supplier of networked
computer systems and services. Digital offers a full range of
computing solutions and systems integration for the entire
enterprise -- from the desktop to the data center.
####
11. Sales Update Draft - DECmpp FORTRAN article
DIGITAL ANNOUNCES DECmpp FORTRAN COMPILER FOR DECmpp 12000 SYSTEMS
Ram Appalaraju
DTN 223 - 4170
MLO1-3/B11
=====================================================================
| |
| o FORTRAN Compiler for DECmpp 12000 Massively Parallel |
| Processing Systems |
| |
| o FORTRAN 90 Array Extensions |
| |
| o Complies with FORTRAN, ANSI X3.9-1978, |
| |
| o Optimizes code for size of DECmpp 12000 system to deliver |
| maximum performance |
|
| o Available in February 1992 |
| |
=====================================================================
Digital announces DECmpp FORTRAN for developing massively parallel
processing applications using DECmpp 12000 systems. DECmpp FORTRAN
is an implementation of full language FORTRAN-77 with FORTRAN 90
array extensions. The compiler complies with American National
standard for FORTRAN, ANSI X3.9-1978. The product is available for
ordering now and will be shipped in February 1992.
PRODUCT DESCRIPTION
DECmpp FORTRAN is a layered software on RISC ULTRIX operating system
and makes DECmpp 12000 Series systems entirely transparent. The
DECmpp FORTRAN optimizing compiler automatically separates scalar and
parallel code, accesses different functional units as needed, and
integrates all communications, and I/O operations. Scalar code is
executed on the DECstation 5000 workstation and parallel code streams
are executed on the DECmpp 12000 Data Parallel Unit. The application
program executes with maximum efficiency because the compiler
optimizes data placement and minimizes data movement. Additionally,
the built-in virtualization allows very large data structures to be
defined and allocated over a smaller physical processor array.
DECmpp FORTRAN FEATURES
DECmpp FORTRAN includes the following features to ease the
development of data parallel programs:
o Functions for array operations.
- Arrays as first class objects
- Assumed shape arrays
- Array sectioning with subscript triplet notation and
vector-valued subscripts.
- Array constructors
- Array assignments
- WHERE statement and construct
- FORALL statement
- Array expressions
- Automatic arrays
o Intrinsic functions to enhance array operations in the
following categories:
- Inquiry functions.
- Elemental functions
- Transformational functions
o WHERE statement
o FORALL statement
o Attribute specification in the type declaration. Supported
attributes are PARAMETER, DATA, INTENT, ARRAY, OPTIONAL, and SAVE.
o Control structures which include a CASE construct, a DO WHILE
construct, and extensions to the FORTRAN 77 DO construct.
BENEFITS
o Delivers high-performance by optimizing user programs for
Processor Element array size transparent to users.
o Fully supported by DECmpp Parallel Programming Environment.
o Supports DECmpp Programming Language calls for programming X-net,
and Global Router.
o Supports DECmpp Image Processing Library using DECmpp Programming
Language.
PRICING AND ORDERING
DECmpp FORTRAN is a separately orderable layered product for the
DECmpp 12000 systems. The SPD number for this product is 36.38.00.
USLP pricing is as follows:
DEVELOPMENT SYSTEM DESCRIPTION
QL-XT6A8-AA $10,000 Software License
QA-XT6AA-GZ $432 Documentation
QA-XT4AA-H8 $335 CDROM Media
QA-XT4AA-H5 $225 TK50 Media
DECmpp FORTRAN is distributed along with other DECmpp software
products in the same media.
AVAILABILITY
February 1, 1992
REFERENCE
SALES UPDATE issue September 16, 1991. Vol. 23 No. 4
DECmpp 12000 Opens Massively Parallel Computing
11. How To Access DECmpp User Documentation
---------------------------------------
From: RDVAX::RODON "Rob Rodon, MLO1-3/B68; 223-2715 11-Dec-1991 1223" 11-DEC-1991 12:29:20.45
To: @DECMPP_ALL.DIS
CC: SUE_D,RODON
Subj: DECmpp FT1.0 Documentation Available
Hello everybody...
I am pleased to announce that the DECmpp FT1.0 doc set is complete
and available in both hardcopy (on RDVAX::) and online (on both
MPSG:: and RDVAX::) formats.
--------------------------------------------------------------------------------
The hardcopy PostScript files are located on:
RDVAX::SYS$PUBROOT:[PUBLIC.MPSG.DECMPP_DOCS]
There is a file named AA_DOC_DESCRIPTIONS.TXT which lists each
manual title and the associated PostScript file.
For those of you who are not familiar with VMS, this is how you
can print out a manual:
1) Enter SET DEF MPSG$PUBLIC at the prompt.
2) Enter SET DEF [.DECMPP_DOCS] at the prompt.
3) Enter DIR to get a directory listing.
4) Enter TYPE/PAGE AA_DOC_DESCRIPTIONS.TXT to read the file.
5) Select the manuals that you want to print.
6) Enter this command for each manual:
PRI/NOTI/QUE=[printername]/PARA=(DATA_TYPE=POST,SIDES=2) filename.PS
I do not recommend printing these manuals on LN03 printers. They
take a long, long time to finish.
Thanks,
Rob
|
| D I G I T A L
- ____ ____ ____ ____
- / / / /___/ /____ / __
- / / / / ____/ /____/
Innovative Systems ...
... for Innovative Customers
MPSG NEWSLETTER #7
Massively Parallel Systems Group - February Update
***Digital Internal Use Only***
Attached is the seventh update from MPSG on Digital's MPP program and
plans. Please let Darlene Camuti (RDVAX::CAMUTI) know if you would
like to be deleted from this distribution list or if you know of
someone who should be added.
Contents
--------
1. Quick status and update on DECmpp
2. Product positioning
3. Creative DECmpp selling
4. DECmpp training
5. DECmpp applications
6. Upcoming DECmpp events
7. Benchmarking
8. DECmpp demos
9. Data Parallel Research Initiative
10. MPP consultants
A1. High Performance Fortran Alliance meeting notes
A2. DECmpp contacts
1. Program Summary
---------------
DECmpp is Digital's entry into the massively parallel
processing market. The DECmpp 12000 is based on a
single-instruction, multiple data (SIMD) architecture. The
products, prices and availability were announced on August
20, 1991. These products are currently on Digital's
corporate price list, carry the Digital logo and are serviced
by Digital worldwide.
MPP systems offer both the maximum computing performance
available and the maximum performance at a given price
level for applications involving large arrays of data.
Examples of these applications include database searching
and sorting, computational chemistry, automobile design,
statistical analysis, text retrieval, signal and image
processing, financial analysis, and seismic processing.
Digital's strategy is sell midrange MPP systems which
integrate into the customers' computing environment. This
integration is accomplished by extending NAS and the
Cohesion toolset to embrace our MPP systems. As part of
Digital's commitment to MPP we have assembled a team of MPP
experts to extend Digital's CASE toolset to apply it to MPP
software development and to port applications to DECmpp
systems.
Our DECmpp platforms range from 1,024 to 16,384 processors
per system and are priced from $250,000 to $1.5M. Peak
performance of the large system is 1.3 Gigaflops.
==========================================================================
DECmpp Fortran is out!
January 20, 1992 sales update contains the article announcing the
DECmpp Fortran V1.0. This is Digital's first product based on
the Fortran 90 standard. The SPD for this software will be in the
SPD database by Feb 1, 1992. Draft copy of all the DECmpp software
product SPDs are in the RDVAX::MPSG$PUBLIC:[.SPD] directory.
Digital is also leading the effort on a High Performance Fortran
building upon Fortran 90. Read more about this in Bert Halstead's
trip report later in this newsletter.
==========================================================================
New MPSG Marketing staff
Mike Anuta joins MPSG as Technical Marketing Consultant
responsible for DECmpp business applications and
opportunities in Washington DC government accounts. Mike
brings 11 years experience in image processing at NASA, the
Defense Intelligence Agency and Digital Equipment.
Mike Fishbein joins MPSG as Product Manager reporting to Ram
Appalaraju. Mike has been with Digital since 1986, in
Software Services as Project Leader, and most recently as
Project Manager at New Ventures Group. Mike is responsible
for DECmpp 12000 products (Hardware and Software).
==========================================================================
Coming attractions:
We are currently evaluating the possibility of offering a
smaller scale version of the DECmpp system, which would be
less expandable and priced starting from $125,000. If you
have definite prospects for this class of system please call
your MPSG regional sales marketing manager listed in
appendix A2.
2. Product Positioning
-------------------
Now that Digital sells Cray's YMP-EL and is preparing
for the introduction of Alpha systems, it is important to
understand how to position DECmpp systems among RISC and
vector products.
DECmpp systems offer the highest performance for a given
price level for applications that map well to MPP. Current
performance is up to 1.3 GF. Significantly, DECmpp software
is a full member of Digital's NAS and Cohesion product
lines. Its operating system is ULTRIX, it uses DECwindows,
the CASE environment will be moving to DEC FUSE and all
networking is based on standard Digital products. The DECmpp
customers' growth path is within the Digital product line
which will grow to also include MIMD MPP.
Applications which map well to MPP include:
Text retrieval financial modeling
large-scale statistical analysis computational fluid dynamics
structural analysis algorithm research
computational chemistry seismic analysis
image and signal processing reservoir analysis
RISC systems (including Alpha in the future) offer medium
high performance over the widest range of applications,
especially those which are less data-intensive. RISC offers
higher performance than MPP for applications which are less
data intensive such as: software development, electronic
mail, office automation, personal productivity, MRP and
numerically-intensive applications that do not involve
arrays of data. Performance of high end Alpha systems (based
on EV-4 to be announced next calendar year) will be in the
range of approximately 70-150 MFLOPS per processor.
VECTORS offer an attractive compromise between highest
speed (of MPP) and the widest range of applications (of
RISC). Many scientific applications have already been
optimized for vector processors. Data-intensive applications
which do not fit well into the MPP application categories
listed above may fit vector processors such as the VAX
6000VP and VAX 9000VP.
The YMP-EL will be a good fit for customers who require full
Cray YMP compatibility and need to run a range of highly
vectorized UNIX applications. Maximum performance of the
YMP-EL is 133 MFLOPS per CPU, up to four CPUs.
3. Creative DECmpp Selling
-----------------------
We have received feedback from sales reps who have found that
DECmpp provides a refreshing entry into accounts requiring
performance previously beyond Digital's capabilities. Since
DECmpp runs ULTRIX and is fully compatible with NAS it helps
them sell the full range of Digital solutions.
We've also found that customers often want to try out these
innovative solutions at their sites prior to their purchase.
MPSG is now offering a "try and buy" program for a limited
time. If you feel you have a qualified DECmpp prospect who
needs to verify the DECmpp solution prior to placing a purchase
order, please call your MPSG regional sales marketing manager
(Appendix A2) to see if our "try and buy" program is
appropriate for your customer.
Qualifying Your Customer For "Try and Buy"
Target DECmpp customers typically have 1-3 large, time
consuming computer applications which require hours or days
to execute on traditional systems. Qualified customers have
a funded project ($250,000 to $1.5M) and either intend to
use or test a DECmpp application code or they intend to
develop their own MPP software. These customers must be
willing to state their acceptance criteria to be met
in order to place their order.
DECmpp Sales Tools
------------------
DECmpp PRESENTATIONS
MPSG public directory (RDVAX::MPSG$PUBLIC:[.PRESENTATIONS]) now has
several presentations newly created for DECmpp sales and marketing
training. The presentations are all in postscript and ASCII formats.
CUSTOMER_PRESENTATION.PS;2
CUSTOMER_PRESENTATION.DOC;2
DATA_PAR_PROG_AND_DEVEL.PS;1
DATA_PAR_PROG_AND_DEVEL.DOC;1
DECMPP_DATA-PARALLEL_COMPUTING_OVERVIEW.PS;1
DECMPP_DATA-PARALLEL_COMPUTING_OVERVIEW.DOC;1
DECMPP_PRODUCTS.PS;2
DECMPP_PRODUCTS.DOC;1
DECMPP12000_SYSTEM_ARCHITECTURE.PS;1
DECMPP12000_SYSTEM_ARCHITECTURE.DOC;1
INTERNAL_PRESENTATION.PS;1
INTERNAL_PRESENTATION.DOC;1
Additional Sales Tools
----------------------
o DECmpp info sheet (EC-F1545-15)
These info sheets are orderable through Northboro
Publishing and Circulation Services.
o Digital High Performance Computing brochure (EC-E1145-57)
o DECmpp configuration guide
(RDVAX::MPSG$PUBLIC:decmpp_configuration_guide.txt)
o Notesfile - RDVAX::DECMPP
o Hotline mail account - RDVAX::DECMPP
o Product manuals -
in the public directory: RDVAX::MPSG$PUBLIC:decmpp_docs.dir
o Demos - available on request, software available on
RDVAX::MPSG$PUBLIC:decmpp_demos.tar
4. Training
--------
ANNOUNCEMENT OF DECmpp 12000 Customer Training
First DECmpp 12000 customer training is scheduled for March 3 - 6,
1992 (at Washington D.C.) Digital Educational Services Training
Center at Landover, MD. The part number for the course is
EY-K018E-L0. Please include the training in the DECmpp 12000
quotations. Subsequent customer training courses will be offered in
different geographies. The training will also be given at customer
sites. For registration information for March 3-6 training please
contact Digital Educational Services directly. For information on
training customers in different geographies (US, Europe, and GIA),
please contact Ram Appalaraju @MLO (RDVAX::APPALARAJU).
DECmpp Product Training:
MPSG is offering three different courses for supporting DECmpp 12000
products:
COURSE COURSE NUMBER INTENDED AUDIENCE
1. DECmpp 12000 EY-H997E-S0 Field Service/Customer Service
Software Training
2. DECmpp 12000 System EY-H998E-S0 Field Service/Customer Service
Maintenance Training
3. DECmpp System EY-K018E-L0 DECmpp Customers, Prospects,
Software Training Digital Sales Support, PSSL.
March 3-6, 1992. Washington DC
4. DECmpp 12000 Sales Digital Sales, Sales Support,
and Marketing PSSL, High-end Drivers,
Marketing..
For information on courses 1, 2 and 3 please consult VTX or
contact educational services directly. DECmpp 12000 Sales
and Marketing course is being offered local to sales geographies.
If there are eight (8) or more students interested in this
training contact your regional MPSG sales marketing manager
(listed at end of this newsletter) for scheduling convenient dates.
NAME: DECmpp SOFTWARE TRAINING (VENDOR TRAINING)
COURSE CODE: EY-H997E-S0
DATES: Call
DURATION: 4 DAYS
FORMAT: LECTURE/LAB
LIMIT: 10
TUITION: $1,000.00
START TIME: 9:00 A.M.
NAME: DECmpp SYSTEMS MAINTENANCE and DECmpp DISK ARRAY
MAINTENANCE TRAINING (VENDOR TRAINING)
COURSE CODE: EY-H998E-S0
DATES: Call
DURATION: 5 DAYS
FORMAT: LECTURE/LAB
LIMIT: 6
TUITION: $2,082.50
START TIME: 8:30 A.M. (Please note earlier start time.)
DECmpp 12000 Sales and Marketing Training
COURSE DESCRIPTION
DAY 1
FOCUS: Sales and Sales Support.
- Introduction to MPSG and goals of training
- MPP style of computing, Digital strategy and plans
- DECmpp 12000 products
- DECmpp 12000 positioning
- DECmpp 12000 Configuration and Quoting
- Selling Vs. Competition
- MPP Applications
- DECmpp Product Futures
DAY 2
FOCUS: Sales Support, PSSL, High-end Drivers, Consultants
Prerequisite: Attend Day 1
- DECmpp 12000 Architecture
- DECmpp 12000 Hardware Description
- DECmpp 12000 Software
- Details of Parallel Applications
- DECmpp Programming Examples
DAY 3
- Customer Seminars
- 1 x 1 Consulting
5. Applications
------------
The following are off-the-shelf codes available for DECmpp:
SAS/TM ACCESS (CHISQ Ltd)
This software provides the large-scale statistical analysis
capabilities of Table-maker integrated into the popular SAS
statistical software package. It provides very high speed
statistical database functions such as data correlation,
averaging and histogramming. The package effectively
replaces mainframes from IBM, ICL, CDC which are currently
being used for statistical analysis. The developer, Ian
Taylor (reachable at dtn 849-3614) located in Reading,
England, is available to provide presales support and
customer consulting for this application.
ELEXIR (Third Eye Software, Inc.)
Elexir is a toolkit for building information retrieval
systems which achieve unprecedented performance for
similarity searches of unstructured text in applications
including legal and financial services, wire information
services, message routing and customer support. Elexir on
the DECmpp 12000 is capable of searching 8 million documents
per second.
Radioss (Mecalog)
Radioss is a car crash simulation code which uses the finite
element method. It achieves 1/3 the performance of a YMP-2E
processor on the DECmpp 4K processor system. This code will
be shown running on the DECmpp system at the Paris
supercomputing show in February.
Codes in process
----------------
The following is a sample of the most promising codes to be
ported to DECmpp systems over the next 3-9 months. We'd be
glad to work with customers who are interested in using
DECmpp systems to run these codes immediately.
CHARMm (Molecular Simulations, Inc)
The CHARMm software package models the static and dynamic
behavior of molecular systems. The code uses empirical
energy functions to compute energies, find minimum-energy
conformations, determine the energy cost of deformations and
simulate vibrational motions. The code is applicable to
small organic and drug molecules, peptides and proteins,
nucleic acids and other biopolymers, and a variety
of synthetic polymers and inorganic materials.
FIDAP (Fluid Dynamics International, Inc)
FIDAP uses the finite-element method to simulate classes of
incompressible fluid flows. The code provides a range of
possible analyses, including isothermal and non-isothermal,
Newtonian and non-Newtonian flows; turbulent flows; free,
forced or mixed convection flows; periodic; separating or
recirculating flows; swirling flows; creeping flows; flows
in rotating frames of reference; flows with a free or moving
surface; and surface tension, gradient-driven thermal flows.
The following codes are under investigation (for short term
conversion, as feasibility is verified).
o Amber
o Gaussian 92
o Xplor
o AVL/FIRE
o Mathcad
o IMSL
o DYNA3D
o AVS
6. Upcoming DECmpp events
====================================================================
DECworld UPDATE
MPSG will be participating at DECworld 92 at the High Performance
Technical Computing Center, and the Innovation Center. DECmpp 12000
system will be displayed at both these places. There will be four
demo stations on the exhibit floor demonstrating brand new
applications and ISV products. Also, MPSG staff will be delivering
session presentation every day. MPSG technical and Marketing staff
will be available throughout the event to help consult and close
business. We expect to show our forthcoming
generations in at a private sandbox suite and will be
prepared to host customer visits and deliver PIDs in the
suite. Stay tuned for further information.
DECmpp Event Schedule
---------------------
February 4 GIA VAX Champions Training
February 12-14 Supercomputing Europe, Paris
March 11-18 Hannover Fair CeBIT '92 (will display DYNA3D)
April 6-8 Data Parallel Research Initiative Symposium, Boston
April 22-24 Supercomputing Japan
Apr 27-May 15 DECworld - including DECmpp sessions and demos
May 4-8 DECUS - Atlanta (DECmpp talks and demonstrations)
7. Benchmarking
------------
The MPSG group has allocated resources for providing performance benchmarks
for potential customer applications. We are committed to dealing with the
benchmark requests in a time-critical fashion. All benchmark requests
must go through one of the MPSG Regional Sales Marketing Managers (see contact
list at end of this newsletter). The benchmark efforts will
be coordinated by Ajit Agrawal (DTN 223-2383, RDVAX::AGRAWAL).
The benchmark code can be sent either on a TK50/tar format media,
or by electronic mail. The request should be accompanied by:
1) The name, phone number, and email address of:
a) Digital Sales contact
b) Digital Sales Support contact
c) A technical contact from the customer
2) Expected date of closing business
3) Date by which the benchmark is needed
4) Expected configuration of the DECmpp machine for which the benchmark is
requested
5) Performance expectation of the customer/salesperson.
We will provide a preliminary evaluation of the benchmark application
within 2 days of receiving the benchmark. The code provided must run
on the DECstation 5000, during the benchmarking process we will
optimize for DECmpp The evaluation will consist of the following
information:
a) An estimate of the performance for the benchmark.
b) The number of man-weeks required for demonstrating the above performance.
c) The turnaround time by which the MPSG will be able to complete the
benchmark.
At the end of this preliminary evaluation phase, plans for further action will
be made in consultation with your sales team.
8. DECmpp demonstrations
---------------------
This update on DECmpp demonstrations is provided by George Anagnostou.
Summary of Current Demo Status:
1) RESIDENCE: All demos reside on the DECstation 5000 named "mpsg"
under the directory "/usr/mpsg2/georgea/mpdemos/ds".
The executables can be invoked on a DECmpp 12000 named "mpdemo" which
is aware of all "mpsg" file systems. A subdirectory called "scripts"
gives a short description of each demo and how to run it.
2) REQUIREMENTS FOR USAGE: To ensure smooth execution of
the software, the environment variable DISPLAY should be set
to your local workstation name:
setenv DISPLAY local_workstation_name:0.0
Furthermore, the session manager should update its security
list to include the system "mpdemo" so that it
can open X-display terminals on the local host. The local
display could be any X-device, but some of the demos
look better in color, so a color terminal is more desirable.
Demos:
I) VOLUME RENDERING: A very nice demo has been
provided by University of North Carolina on ray
tracing and volume rendering on a collection of images
obtained by Nuclear Magnetic Resonance Imaging. It is a 3-D
representation of a human head and the data
exists in two pixel (or voxel, rather) sizes: 128 and 192;
for anything less than 4k machine, the smaller data set is
recommended. Once the initial window is
opened, the image can be manipulated with the left and right
mouse buttons. Available functions are, rotation, translation,
light source orientation, and opacity levels. The executable
resides under "unc2.0/vol" and is invoked by typing "vol 128.im"
or "vol 192.im".
II) COLOR SORTING: Another impressive demo has been
contributed by U of North Carolina as well,
where a particular color pattern is searched based on an
initial random distribution. The intermediate color states are
displayed on the graphics window as they
are updated by the back end. There are three different
executables based on three different sorting algorithms. A nice
visual effect is obtained by running all
three executables to see the particular color patterns picked
by each algorithm. This demo resides under "unc2.0/vis" and is
invoked by typing fla, xnet, and srm for
the three algorithms, respectively.
III) RADAR TRACKING: A not-so-exciting application of
object detection and tracking from a
collection of raw radar data across a pre-defined area of
interest. The parallel algorithm searches
through the data, filters out the noise,
determines the real objects, and tracks them. There
are three different executables in this directory with different
domains searched through in each case. The executables reside in
"tracking" and run by just calling the name of each executable.
IV) SLALOM: A non-graphical demonstration of the time
it takes for the solution of linear equation based on the radiosity
solution benchmark developed by John Gustafson. This application
should be avoided unless there is a specific request.
V) MPIPL: This is a demonstration of image filtering
using the Maspar Image processing
library. The application processes various satellite images of
cities around the world and applies FFT, and
other filters to them. For more impressive
results the demo should be run from a color terminal and the
mouse should be clicked on the graphics
window for correct color representation. The executable
resides in "mpipl" and is invoked by saying "mpipl -p city"
where city can be any of the following: marseille, paris,
riyadh, dakar, land, saturn etc.
VI) LINPACK: Not a flashy demo, and should only be run
if there is a specific request for it.
As the name implies, this executable solves linear systems of
equations and displays the time it took the DPU to complete the
job. The demo resides under "linpack" and is run by typing lups.
VII) MANDELBROT. This is a nice implementation of the
search for a Mandelbrot set on a 400x400
pixel square cutaway of the plane of complex numbers. The
application starts from some default position of the cutaway and
assigns a complex number to every pixel
which in turn gets assigned to every processor on the DPU.
Once the graphics window is painted, the user can change the
position of the square or zoom into the plane by clicking
the left and middle mouse buttons, respectively.
The right mouse button is used to reset the whole plane. The
time between repaints on the display window is the time it
took the DPU to compute which of the 160000
complex numbers are in the set and which are not. The demo
resides in "mandelbrot" and is invoked by typing "mb".
IX) TIN-CAN ON TURN TABLE: The title speaks for itself.
This application makes use of Marc Picart
image processing library which is still under development
in collaboration with Thompson CSF of France. The graphics
window does not make use of color. It displays a can
positioned on a turn-table which is
rotating. The location of the executable is under "crl".
X) ARCO MIGRATION CODE: This is a rather dull
application of seismic migration. The
display window shows the propagation of an acoustic wave through
a volume. The discretization is a grid of 256x256x256 allocated
at the back end. This demo should be
displayed only to customers related to the Oil & Gas industry.
The code will be available starting Feb. 7th under the
directory "arco".
Work in Progress
----------------
The following is a list of demos that are planned for the DECmpp
system in the future. Some are more long-term opportunities
than others; the individual status for each one is given below:
I) SAS - TABLEMAKER: This is a nice application of
database searching executed in
parallel. Table-maker is a superset of SAS (Statistical Analysis
Software) which runs on the DPU and can therefore, be considered
as a turbocharger to SAS.
The demo unfortunately, requires the installation of SAS
which in turn needs a license key to execute. The demo copy of
the software expires in a month at which point we will
decide to extend the license.
II) RADIOSS: A crash simulation package by Mecalog in
France. The port to the DECmpp system is
almost complete and will demonstrable by the end of January.
The demo data set will most likely be the crashing of a steel
bar.
III) ELEXIR: A text retrieval software application
provided by Third Eye Software. Robin Young is
the contact person for the port.
IV) AVS: A superb visualization package where some of
its compute-intesive modules are ported on
the DECmpp 12000. Discussions are currently undergoing
for New Mexico State University to do the port which will most
likely to complete by the end of summer.
V) CAT-SCAN RECONSTRUCTION: An application from
Washington University in St. Louis. Discussions are undergoing
for them to get a DECmpp system. Their area of expertise is in
image processing for medical applications.
VI) FIDAP: A fluid dynamics package by FDI in Evanston,
Illinois. Their main linear equation solver
will be running on the DPU with all the graphics post-
processing performed at the front-end. This is scheduled to take
place around the month of April.
VII) DYNA3D: A port performed by Joseph Pareti in
Germany and scheduled to complete by end of February. The demo
problem will most likely be a complete car crash simulation.
9. Data Parallel Research Initiative
---------------------------------
The Data Parallel Research Initiative is a Digital External Research
Program to promote and fund research and software development in the
field of massively parallel computing. As part of DPRI Digital funded
research institutions worldwide and provided Digital front end systems
to SIMD systems from MasPar and Thinking Machines. Digital will be
sponsoring a symposium this April in Boston at which DPRI researchers
will present their results.
Overviews of the research work of each institution is available
on-line in RDVAX::MPSG$PUBLIC:DPRI.MEM
Institution Area of Research
----------- ----------------
Florida State University Theoretical High Energy Physics,
Experimental High Energy Physics,
Ocean Modeling, Economics,
Theoretical Condensed Matter Physics
and Materials Science, the Human Genome
Initiative, Chemistry,
Workstation Environment
Texas A&M University Simulation of petroleum reservoirs,
Modeling of vehicular traffic flow,
Neural networks
Harvard Univ. Parallel Implementation of Dynamical
Systems for Intelligent Machines
McGill University Parallel-Iterative Algorithms for
Image Reconstruction Analysis, and
Interpretation
Toyohashi University of Tech. Data-Parallel Extensions to a Common
Lisp System
Ecole Normale Superieure de Lyon
Implementation of a general purpose
dictionary machine on a data parallel
computer
CSIRO Parallel Algorithms for Interactive Analysis
and Visualization of Spatial Data
University of Bath Algorithms and Data Structure for Symbolic
and Numeric Computation
The University of Akron Software Automation for Advanced Parallel
Computers
Bellcore Selected Parallel Sparse Matrix Algorithms
Universitaet Stuttgart Programming Environments for Massively
Parallel Programming and Neural Networks
Purdue University Automatic Generation of Massive Parallelism
The Johns Hopkins University Data Parallel Algorithms for the
Numerical Modeling of Semiconductor Devices
Boundary Contour System (BCS)/Feature
Contour System (FCS) Model for Smart
Robotic Sensors
A Parallel Algorithm for Sensor Array
Signal Processing
University of Vienna Development of programming environments
for parallel systems
The Tokyo University, Massive Parallel Database Processing/Massive
Institute for Industrial Parallel Image Processing
Science
Carnegie Mellon University Application of data parallelism to computer
vision
University of Oregon Genetic and Demographic Simulations of
Endangered Species
LOUISIANA STATE UNIVERSITY SIMULATION AND VISUALIZATION OF COMPLEX
AND A & M COLLEGE PHYSICAL SYSTEMS
University of Melbourne Development of Data-Parallel Applications
Utrecht University A Problem Solving Environment for Ocean
Circulation Simulation on Massively
Parallel Computers.
University of Washington Data-Parallel Systems, Applications, and
Algorithms
Princeton University Parallel algorithm development for (a) Monte
Carlo simulation of disordered materials and
related optimization problems and
(b) Mathematical modeling of semiconductor
device structures.
Florida Atlantic University Monte Carlo simulation of phase transformation
in two-dimensional and layered systems
IRISA, Campus de Beaulieu Portability parallel software development tools
INSA, LMFN Modeling of Turbulent Combustion in Supersonic
Flows- Application to Hypersonic Propulsion
9. DECmpp consultants
------------------
MPSG has begun to line up outside consultants who are available to help
Digital and our customers port and optimize their applications on DECmpp
systems. Below is a list of some of these consultants. Please contact your
regional MPSG Sales Marketing Manager (see back of this issue) for more
information on this topic.
submitted by Bill Celmaster - January 30, 1992
Geoffrey Fox (Syracuse)
Matthew Whitten (U.T. Austin)
John Campbell (private)
Joel Saltz (ICASE)
Rob Schreiber (RIACS)
Rich Brower (BU)
Gary Montry (private)
Alan Edelman (Berkeley)
Dongarra/ORNL (U Tenn)
Oliver McBryan (U Colo)
Kuch and Associates (KAI)
Henno Allik (BBN)
Grover Cleveland (private)
Joel Mannion (private)
Scientific Visualization Associates
Scientific Computing Associates
Visionary Systems
A1. High Performance Fortran Alliance Meeting Notes
-----------------------------------------------
Digital is helping to set the standard for high performance or data
parallel Fortran. This effort is building on Fortran 90, with
particular attention paid to a set of directives for optimizing codes
to MPP architectures. Ken Kennedy of Rice University, one of the
foremost authorities in compiler research, accepted our invitation to
chair this effort.
Digital's view of the standard was presented informally at
Supercomputing '91, and the attached trip report shows that Digital's
leadership effort is being very well received.
As with all material in this newsletter, this report is
strictly for Digital's internal use only. Please feel free
to contact your MPSG regional sales marketing manager to set
up discussions about high performance Fortran with your
customers.
29-JAN-1992 10:14:17.69
From: CRL::HALSTEAD "Bert Halstead"
To: mpsg::ganley, rdvax::nilsson
CC: tle::harris, tle::whitlock
Subj: trip report -- High Performance Fortran Forum, 1/27 - 1/28/92
High Performance Fortran Alliance Meeting Report
------------------------------------------------
This is a quick report of my impressions of the inaugural High
Performance Fortran Forum meeting, held in Houston near Rice
University on 1/27-1/28. About 100 people attended the meeting,
including representatives from every well-known massively parallel
hardware vendor (for example, Intel, Cray Research, Thinking Machines,
MasPar, IBM, Motorola, nCube, Alliant, Kubota Pacific, Kendall Square
Research, and Convex were all represented; but AMT, Wavetracer, Tera,
and Sun were not), many universities, many government labs, and a
surprisingly large number of Europeans (I'd say there were about five)
who had traveled all the way to Houston for the meeting. The subject
of this meeting was clearly of great interest to a wide community.
The agenda for the first day featured 30- to 60-minute presentations by
a whole range of parties: Digital, Cray, the Rice University Fortran D
project, Hans Zima of the University of Vienna, Jorge Sanz of the IBM
Almaden research lab, Convex, and Thinking Machines.
With such a huge number of attendees, and such a plethora of
presentations, there was a great risk of the meeting being completely
unfocused and unproductive. However, given these initial disadvantages,
the meeting went as well as anybody could have hoped, both for the
prospects of converging on a de facto High Performance Fortran standard,
and for Digital's role in defining that standard. This was due in no
small part to the strong leadership exercised by Ken Kennedy (the
convenor of the meeting) and to the very positive reception accorded to
the presentation given by our own Dave Loveman.
While the number of presentations given might suggest that the HPF Forum
would be faced with more proposals than it could succeed in dealing
with, the truth was not so grim. Only three speakers -- Dave Loveman,
TMC's Guy Steele, and Chuck Koelbel of Rice University -- addressed
themselves directly to the topic of what language definition the HPF
Forum should actually adopt. Digital's proposal was received exactly
the way we intended -- as a minimal base that we could all quickly agree
on, and then go on to debate more controversial matters. The only
serious controversy was over our embracing full Fortran 90 (there was a
lot of sentiment favoring including only a subset of Fortran 90, or even
just FORTRAN 77). Aside from minor nits, I didn't hear any substantial
or serious disagreement with accepting our proposals as at least a
subset to build on. And Dave's presentation (which was scheduled right
at the beginning of the meeting) was received extremely well, as setting
just the right tone for the rest of the meeting.
The Thinking Machines presentation, by Guy Steele, was also very good,
and contained a number of good ideas (though generally not as polished
and finished as ours) delivered in the spirit of, "Digital's proposal
will be great if we can just fix these few things." The tone of both his
written and oral comments was very constructive. Chuck Koelbel also
took a very constructive point of view, presenting the "wish list" of
the Fortran D project but also pointing out areas for compromise.
Tom MacDonald of Cray presented their programming model, not
as a proposal for the HPF Forum to adopt, but just to let us know what
Cray is up to. Hans Zima gave a good presentation about Vienna Fortran
but it didn't really appear that he expected the HPF Forum to adopt
Vienna Fortran -- he just wanted us to know about the ideas that he
has been exploring. Jorge Sanz of IBM Almaden gave a presentation of
some of his own research ideas -- not very fully developed yet,
certainly not an "IBM proposal for HPF". Joel Williamson of
Convex gave a presentation which consisted of pointing out the things
he didn't like about each of the other proposals, without offering any
proposal of his own. The second day of the meeting began with Ken
Kennedy taking the floor to discuss how the HPF Forum process should
proceed from this point onward. He stated his sincere desire to have a
small working group, consisting of members who were willing to commit
to attending nine or so meetings over the next 12 months. He said he
wanted to limit attendance to only one representative per company or
organization. (It appears that the working group will consist of
about 30 people -- large, but manageable if Ken sticks to his activist
style.) He mentioned the Digital proposal several times as a great
starting point, and complimented Dave Loveman both publicly and
privately for setting out so clearly the issues we should focus on.
(Dave's slide full of boxes showing all the different Fortran subsets
came in for particular praise.)
In the break that followed, Ken indicated that he wanted Dave to lead
off the next HPF Forum meeting by again reminding everybody of what the
issues are. Ken also wants Dave, Guy Steele, and maybe some other
people to get together before the next HPF Forum meeting and work out
the details of (ideally) a joint proposal to put before the Forum. As a
measure of Ken's commitment to this process, he wants to schedule at
least one such meeting at a time when he can fly up to Boston to
participate in it himself.
There are still many things that could go wrong with the process, but it
is very encouraging that the three main purveyors of serious proposals
-- Digital, TMC, and Rice -- are all so close to agreement and appear
willing and able to work together. It is also very encouraging to see
Ken Kennedy's leadership talents and interest in making this project
successful. Moreover, to have a figure as respected and impartial as
Ken Kennedy choose to single out Digital's contribution to the Forum for
as much praise as he did, before the distinguished and influential
audience that attended this week's meeting, has to help a lot in
convincing people that we are serious about Fortran for massively
parallel processing and we do know what we're doing. Finally, Ken
clearly sees Digital's role as central in the evolution of the HPF
specification, and this also meets our objectives.
-Bert
A2. DECmpp Contacts
---------------
MPSG Regional Sales Marketing Managers:
Eastern U.S. Steve Harrington, 223-3435, RDVAX::HARRINGTON
DC Government Mike Anuta, 341-5277, HMRJKS::ANUTA
Western U.S. Eric Tyberg, 521-6860, USWRLS::TYBERG_ER
Central U.S. Ron Renaud, 223-4161, RDVAX::RENAUD
Pan Europe Steve Zagame, 858-1067, CCIIS1::ZAGAME
United Kingdom Simon Cole, 899-3028
Germany, Austria,
Switzerland Jochen Krebs, 786-1408, MTS$::"unt::jochen KREBS"
France Yvan Bantoure, 858-1397,MTS$::"FGT::YVAN Bantoure"
Nordic countries Magnus Persson, 876-8000, mts$::"soo::magnus persson"
South Pacific, Asia Greg Madden @SNO, DTN 730-5454, (61-2-561-5454)
European product mgr Bob Pattison, 899-5734, CHEFS::PATTISONB
Japan Hiroshi Mizoguchi, 03-989-7340, TKOVOA::mizoguchi_h
o Service contacts Diana McClure, 264-0883, MVDS02::MCCLURE
Mike Carlton, 264-5740, MVDS02::CARLTON
o Additional Contacts --
Mike Fishbein, Product Manager, RDVAX::FISHBEIN, 223-2209
Ram Appalaraju, Program Manager, RDVAX::APPALARAJU, 223-4170
Bob Nilsson, MPSG Sales & Mktg Manager, RDVAX::NILSSON, 223-1715
Charlie Wilson, MPSG Group Manager, RDVAX::WILSON, 223-2867
DECmpp Application Consultants
Bill Celmaster, 223-2173, RDVAX::CELMASTER
Ajit Agrawal, 223-2383, RDVAX::AGRAWAL
George Anagnostou, 223-2209, rdvax::anagnostou
Bob Monroe, 531-5271, rdvax::monroe
o Order processing contact -- DECmpp products are available
for quoting in the AQS system. For help with quoting and
processing orders, contact Lisa Peloquin, DTN 223-1935,
RDVAX::PELOQUIN.
o In-field support people who have received DECmpp training
Over sixty in-field sales support people have been trained
on DECmpp. Give the MPSG support person assigned to your
geography a call for help in assembling a team for your
sales opportunity.
|