Conference ulysse::rdb_vms_competition

403.0. "Restoring a BIG database" by HGOVC::MICHAELWAN () Thu Aug 17 1989 07:11

  **This note is cross posted in RDB and RDB_VMS_COMPETITION notes files**
    
    I asked some similar questions 5 months ago and didn't get too many
    answers. Since there have been a lot of developments on  Rdb
    applications in the field, I guess it is the appropriate time to
    re-visit those questions again.

    The basic problem we are facing with is to support a 30+ GB Rdb
    database. The major problem so far is the recovery issue. I have also
    been told by product management that multi-streamed RMU restore will
    not be available due to data integrity problem and is technically
    impossible to do.

    Therefore, the restoring time of the database will basically limit by
    the speed of one single tape drive. To restore a 30+ GB database will
    then take 2 days !!! I don't think any customer can accept this as a
    recovery time.

    One alternative is to break the database into several smaller ones. We
    have ruled out this due to other reasons. (lacking of 2PC is one
    reason.) BTW, didn't we claimed that we can support a 50GB Rdb
    database. Use of volume shadowing is a must, but that only covers the
    disk failures.

    We are planning to use VMS backup/restore to replace RMU backup/restore
    since it is the only way we can think of to speed up the recovery
    process. (I am aware of the drawbacks of VMS backup/restore such as no
    incremental backup (in the DB sense) and online backup....)

    Questions: -

    1) Had anyone in the field faced the same problem and got some good
    solutions?

    2) If I use VMS backup/restore, assuming there's no bottle neck, then
    the backup and restore time is only limited by the number of tape
    drives and tape drive speed. Probably, we can backup or restore the
    whole database within 4 hours if their are not CPU bottle neck, bus
    bottle neck, controller bottle neck, CI bottle neck ....

    In reality, what will most likely be the bottle neck and how will that
    affect the performance? Where can I find more information about this to
    do a more precise calculation? (Any one from the VMS or performance
    group?)

    3) The combination of extra HSC and disks to support volume shadowing
    is so costly that we are afraid that the customer will go back to the
    IBM solution. The reason they have chosen VAX and Rdb is because they
    like VAX and relational DB technology.

    Can IBM supply a solution with DB2? What is the size of the biggest DB2
    database in the world.

    4) What size of a database does IBM claim that DB2 can support? 50 GB?
    What is reality?

    5) Can they restore a DB2 DB in multi-stream? (and backup in
    multi-stream?)

    6) If not (5), can they get away with it by 2PC and multiple smaller DBs
    so that single-stream restore is not a concern?

    7) How fast can they back up and restore? (examples?)

    8) In general, who in the industry HAVE successfully supported a
    relational database of 30+ GB? How can they manage the backup/restore
    problem? 

    Thanks.
    Michael

    Try the US Censous Bureau, Washingtion. The have the largest database in the
    world, and they just went to Rdb. I read the brochure yesterday.
    I don't know who is working in that group, but I am sure that they
    have knowledge of how to handle big databases like that. 
                   
    Peter

See the replies in the RDB30B conference