| T.R | Title | User | Personal Name
 | Date | Lines | 
|---|
| 277.1 | Let' Be Reasonable | MDVAX1::DUNCANG | Gerry Duncan @KCO | Thu Dec 15 1988 20:49 | 13 | 
|  |     A couple of thoughts.  Any company in the financial business must
    be concerned about availability and growth.  Our answer to this
    critical customer need is clusters ... pure and simple.  Assuming
    the customer believes that clusters are his protection and  growth
    it only makes sense to have a cluster benchmark as a part of the
    evaluation criteria.
    
    It doesn't matter if one uses a database or flat files, multiple
    disks have ALWAYS been the environment of choice for companies who
    must have maximum performance.  Asking the customer to include that
    approach is, in fact, reality.
    --gerry
 | 
| 277.2 |  | SNOC01::ANDERSONK | My DEBIT/CREDIT performance is lousy | Fri Dec 16 1988 03:25 | 17 | 
|  |     ADABAS allocates files to disks by cylinder, track etc, and so does
    not have dynamic file extension on the database file. Can you suggest
    theat the benchmark includes a performance test on what happens
    when the database has to be extended? There is no doubt that ADABAS
    can be setup to be fast...
    
    Force AIJ to be on for any benchmarking (probably makes no difference
    to ADABASE) - any real system has got to have this. (Our local ORACLE
    customer has been conned into believing that his applications dont
    really need AIJ as the BIJ is good enough).
    
    
    Does your customer want strict relational DBMS? Ask for compliance
    with Codds rules (ADABAS is not really relational, and I have advice
    that all good performance stuff is nearly always from setting up
    data in a non-relational way). Thats not really related to benchmarking
    as such. Do they want degree-3 consistency?
 | 
| 277.3 |  | CREDIT::KELLEY | Confused, I am so.... | Fri Dec 16 1988 15:11 | 14 | 
|  |     Having seen ADABAS on a couple of times, one thing I suggest you
    should keep your customer's benchmark away from is having multiple
    users updating a hot spot in the database.  Since ADABAS has a read
    for update command, they can put a write lock on the record where
    Rdb puts a read lock and then promotes it to a write lock.  Rdb
    therefore gets a lot of deadlocks and rollbacks, while ADABAS continues
    to run.  The kind of hot-spot I am talking about is things like
    a record that contains an order-number where everyone reads, adds
    1 to it, and writes back.  I know that we can do that with an exclusive
    write transaction, but then you have to have logic in the program
    that traps the transaction failure and customers where we lose are
    those who think the DB should handle those.
    
    chuck
 | 
| 277.4 | some things against ADABAS | HSK01::KORVENRANTA |  | Mon Dec 19 1988 13:03 | 29 | 
|  |     
    Hello,
    I have some things in mind what you could use against ADABAS:
    - Reorganization of the whole database has not been working at
      one of our customers at all. They have to delete and create
      each file one by one.
    - If you want to add a new field to a file, there cannot be any
      users in the database who have been using that file during his
      session. No matter if there are no users at the very moment...
    - If you use ADARDA; the network product whith which you can read
      (and update) over network, there is no recovery in that product.
      If DECNET falls down, ADARDA cannot even come up.
    - There is no hashing (at least in that version that our customer
      has). 
    - If you want to recover with backups and protection logs (PLOG),
      you have to make every changes to the database structure by hand.
      The recovery program always stops and says that now you have to
      add the new field ... to the file ... After it you can continue
      the procedure that restores the data (until there is again some
      changes that you have to make by hand..).
      The only time that customer needed to restore the database, even
      that did not work...!
    
    Maybe the newist version solves some of these, that version I have
    not seen.
    
    Regards,
    	P�ivi Korvenranta
    
 |