Conference 7.286::fddi

The DEFZA does not do DMA on transmit.  (ie. DMA is only done for receives)

Can you try the new FDDI Turbochannel adapter?  It has both transmit and receive
DMA, and generally has a more efficient architecture.  It has performed quite
well in the lab.

doug

    
    Doug,
    
    First..thanks for you response....but I/we still don't understand why
    there would be no DMA on the XMIT side of a hi-speed FDDI controller.
    My customer is a highly technical, widely respected national lab and I
    can't go back to them with merely "buy the new adapter to solve the
    problem". They are threatening to broadcast this wide and loud on
    Internet unless they get a straightforward answer.
    Second..I have to beg ingnorance of any new FDDI T-channel controller.
    Is this announced? If not can you indicate where I can get preliminary
    details.
    I greatly appreciate any details you can provide. Thanks !!
    
    Denis

There are two reasons why the first FDDI adapter (the turbochannel 700) did not
have transmit DMA.  The first reason is that it was quicker time to market to not
do transmit DMA (TTM was of primary concern, not performance).  The second reason
is that the west coast folks argued that transmit DMA would not buy the customer
anything.  [I don't think they counted CPU cycles as being important to a 
customer.

The receive DMA engine has several limitations as well.  First, the firmware gets
involved in each and every packet, and second, a minimum of 512 bytes is DMA'd, 
even for 1 byte of data.  This can increase latency for small packets
unnecessarily.

I'm not exactly sure of the status of the DEFTA (the new turbochannel option).
Contact Sharon ONeill for details.

doug

Are there any DEFZA -> DEFTA upgrade plans ?

Stefan

    Stefan,
    
       You should probably send mail to Sharon (DELNI::ONEILL) Oneill regarding
    any DEFZA to DEFTA upgrades.  She is the product manager for all of the
    FDDI Adapters (DEMFA, DEFEA, DEFZA, DEFTA, et al).
    
       - Larry Stefani
         DEFEA Project Engineering

Something about this "highly respected lab" bothers me.  The adapter works
exactly as described.  Why are they upset?  Did someone tell them it has
DMA both ways?  It doesn't, never did, never will, never was designed to.

	paul

The DEFZA was out a long time (a year?) before the VS4000-90 came out, and used
on DECstations under Ultrix, so its design tradeoffs undoubtedly had the fast
MIPS CPU speed factored in.  The early TURBOchannel SCSI disk controllers don't
do DMA in either direction, by the way. 

I think a good trade-in offer is probably the best way to go. 

>The DEFZA was out a long time (a year?) before the VS4000-90 came out, and used
>on DECstations under Ultrix, so its design tradeoffs undoubtedly had the fast
>MIPS CPU speed factored in.

    Correct me if I'm wrong, but some NVAX models are *faster* than
    DS 5000/200 and perhaps DS 5000/240.  Certainly the difference in CPU
    speed is minor.

    My theory: Unix and VMS users tend to have different ways of using
    machines, and different monitoring tools.  Perhaps the high CPU expense
    of driving DEFZA is less visible on Ultrix than on VMS.  And for
    historical hardware reasons, VMS users are more likely to closely
    watch the disappointing progress of their jobs, and to diagnose
    performance killers in the system.  8^)

    I've heard a Digital vice president champion programmed I/O even on
    slow, low-cost VAXes.  "The CPU chip makes a great DMA engine.  Our
    performance study shows that dedicated DMA hardware doesn't help much."
    This argument fails to recognize that some users have multiple tasks
    running at once.
    --gw

    Not that the author of .7 deserves a reply re: the design of the DEFZA
    but I do wonder if he was on the GM gas tank design team?? I find the
    "never designed to " quote most amusing...I'm sure the engineers at the
    "highly respected" lab would also....maybe you'd like to relay that
    message?
    
    Because of the fact that early SCSI disks didn't do DMA as pointed out
    in .8, we made a point to find out if this device did DMA...guess we
    just "forgot" to ask if it was bi-directional.
    
    Also , in reply to .8  what Decstation was faster than the 33 spec
    VS4000 when the DEFZA was first introduced?
    

Well, you originally talked about this as if it were a bug, which of course
it isn't.  It may have been a bad design decision (though the opinions of
several highly respected senior people are, at best, mixed on that).
But your only recourse is to tell the customer "yes, you're right, it doesn't
do DMA".

	paul

    When the DEFZA was introduced it was one of the first FDDI adapters on
    the market. It was less than half the form factor of the closest
    competitor and less than half the price. I don't think we should get
    into a pissing contest about a decision made over 3 years ago to only
    do DMA in one direction in order to complete the product design in
    under 6 months. Thats not going to solve the customers problem. The
    FEFTA hardware has been completed almost a year ago and it has been
    working with a driver for at least 6 months. It will be released with
    an OSF driver in March and a VMS driver in something like May. 
    It is the highest performance FDDI adapter available for sale by anyone
    today. NOTE - it may not be the lowest latency but it is the best for
    moving the most Mbits/sec. Hopefully you can work something out where
    they can trade the antique DEFZA's for DEFTA's. 
    I will try and post the product announcement for the DEFTA as the next
    reply - if I can't I will mail it to this notes originator.