Andy,
Would you please update us again on the issue of mirroring vs RAID for TSM
log and database volumes.  Looking back in the ADSM-L logs, I believe your
most recent recommendation (June, 1998) was to continue using TSM mirroring
w/ Sequential Writes to protect against "partial page writes" which could
be propagated to the other mirror(s) if using operating system or hardware
mirroring.  Have circumstances changed/improved since then?

We are specifically interested in the following circumstances:

1.  An IBM consultant has recommended using SSA RAID 1 or 10 mirroring
    w/ 6225 (4-P) SSA 160 SerialRaid Adapters w/ FastWriteCache enabled
    to protect and optimize performance of our Oracle database.  Can we
    safely use the same to protect and optimize performance of TSM log
    and database volumes, eliminating the need for TSM mirroring?

    (Our current understanding is that FastWriteCache w/ battery-
    protected cache memory will eliminate the "partial page write"
    problem.  Is our understanding correct?)

2.  Has anything improved w/ AIX logical volume mirroring to lessen or
    solve the problem of "partial page write propagation?"

3.  If we must continue to use TSM mirroring, what are the performance
    implications?  Management wants to know how much of a "performance
    hit" we incur by continuing to protect our TSM DB (raw logical)
    volumes w/ TSM vs AIX vs RAID1 mirroring.

    a.  Does TSM optimize page reading across TSM-mirrored volumes?

    b.  Are there any other performance mitigations that TSM-mirroring
        offers that we might not have considered?

4.  Our 50->60GB TSM database is currently TSM-mirrored on 3 pairs of 18GB
    + 1 pair of 9GB SSA disks, each containing a single Raw Logical Volume.
    We believe we know "what we can't/shouldn't do w/ RLV's so that we
    don't damage the TSM volumes" and we continue to believe that RLV's
    are simpler more efficient than JFS files for TSM volumes.

    a.  Has TSM changed since ADSM v3.1 such that we should convert from
        RLV's to use JFS file volumes for efficiency?

    b.  If so, should we define 1 JFS file/disk or another configuration?

    c.  If/When our TSM DB grows 60->70GB are we better off w/ 4 * 18GB
        SSA 160 disks or 8 * 9GB (older SSA) disks.

Thank you so much for your help and consideration!
--
Jim.Owen AT Yale DOT Edu   (203.432.6693)
Jim.Owen AT Yale DOT Edu   (203.432.6693)