Conference ulysse::rdb_vms_competition

35.0. "Ingres uses mailboxes?" by LILAP::WATERHOUSE () Sat Oct 10 1987 21:04

    I am a resident at a site using rdb and the customer is looking
    very hard at Ingres.  They have recently hired a person in their
    database group with Ingres experience and that person is doing the
    evaluation.  Needless to say the information being generated is
    slanted towards Ingres.  The environment we need to support is
    approximately 400 users running a dispatching service for a railroad
    (e.g. moving freight around the country via rail or trailer).
    
    I have ACMS/TDMS with RDB on the machine now and have prototyped
    a system for them with no problems.  I am very familiar with RDB
    having done several projects with it and some performance consulting
    too.  By the way it is currently a 2 8550 cluster with 6 RA81's
    with the probability of more of each.
    
    My current understanding of Ingres via reading and talking to internal
    people is that Ingres uses a front-end and back-end process for
    every user and if you use graphics and something else you have an
    additional 1 or 2 processes.  I also understand the processes 
    communicate via mailboxes.  This must introduce significant overhead!
    
    I also understand that in the rally type development mode that all
    messages are passed to the backend processes where they are parsed,
    optimized, (compiled?), and executed, finally returning the result
    to the foreground process.  This happens every time and there is
    no concept of a compiled transaction that is re-used in this
    environment.  In the 3rd generation approach (embedded dml), compiled
    transactions are used but you lose the elegent development environment.
    
    Is the foreground /background processing technique still used in
    the embedded DML environment?
    
    Are mailboxes really used for passing messages?
    
    Can they support 400 users in a 3 tps environment or will they bury
    the machines (40 mb each)?
    
    Will the user have to write a lot of code to get the performance
    they require in Ingres and will they be able to tune the application?
    
    I understand Ingres will be going to a server concept ala ACMS but
    they will still have 400 front end proceses and they will probably
    still use mailboxes.  Why not global sections?  I find this hard
    to beleive that is why I am posting this note.  I need help ASAP
    and would appreciate any you can give me?
    

    see note # 86 for some good information on Ingres. Jamey Nordby