Walter,

Yikes.. you made your diskpools RAID5 on NT. NT is slow enough without
having to make it's DISKPOOL RAID5.
I'd go for RAID0 for diskpools and if I had the money RAID0+1 for a bit of
fault tolerance.

I suppose that depends upon how critical you view your diskpools. To me the
diskpool is just a temp. staging area
therefore in the event that I lost a disk, I'd re-do my diskpool and rerun
last night's backup.

If you're doing stuff like HSM then you might want more fault tolerance in
there.

Incidentally my throughput doubled on diskpool perf when I went from RAID5
to RAID0+1.

As far as DB and Logs, I don't think that anyone can afford to be as risky
as RAID0. However again RAID5 would be my last
choice here. I'd rather go with a RAID1. Common sense suggests that devoting
as many spindles as possible to the DB and log will give you the best
peformance i.e. rather than mirror a single 20Gb drive, mirror 5 4Gb drives.
Real world stats may prove me wrong on this but makes sense to me...

Then when creating the Database volumes create multiple volumes per physical
drive. If you've been reading the list a lot of people have pointed out that
a separate thread is allocated for each database volume, therefore the
larger the number of database volumes the greater the throughput. However
there must be some sort of limit to this... .e.g If I had a 10Gb disk volume
which I needed to create database volumes on, I reckon that I'd hamper
rather than improve performance by creating 100 100Mbyte database volumes.
Some experimentation would be required here to find out where the balance
lies.

Just my thoughts...

Nathan








        -----Original Message-----
        From:   Walter Ridderhof [SMTP:Walter.Ridderhof AT MAIL.ING DOT NL]
        Sent:   Monday, September 06, 1999 9:54 AM
        To:     ADSM-L AT VM.MARIST DOT EDU
        Subject:        Re: disk-pool performance

             R. Masa, Ramesh,

             I'm sure the client to volume holds true for direct to tape
backup, as
             a drive per client is assigned, but t'ill now I havent't heard
the one
             on one client->disk stgpool volume strategy. As far as my
knowledge
             goes all diskpool volumes are evenly filled with client data
             regardless the number of clients attached at any one time.
Inside
             story is that ADSM assigns a disk read/write thread per defined
disk
             storage pool volume, so having multiple volumes in a disk
storage pool
             would certainly help write performance for that storage pool.
If your
             running UNIX ADSM assigning a disk per volume wouldn't be a bad
idea.
             Under NT, this depends if your using RAID or not, if not you
can
             follow the UNIX strategy otherwise well, commonsense and a good
bit of
             RAID knowledge will ultimately give you your answer.
             So far for NT and disk storage pools I've always gone for
RAID-5 (3+
             disks) and made as many volumes as there were disks available
(not
             backed by any performance measurements).

             If anyone has some real world experience on optimal ADSM
             DB/log/diskpool configurations I would sure like te hear about
it.


             Walter Ridderhof, Mainland Sequoia


        ______________________________ Reply Separator
_________________________________
        Subject: Re: disk-pool performance
        Author:  "ADSM: Dist Stor Manager" <ADSM-L AT VM.MARIST DOT EDU> at 
INET-1
        Date:    9/2/99 10:17 AM


        It is not recommended to have one large storage pool volume.  As
each volume
        can allow only one client at any given time to write to it, you
should have
        at least the same number of stg-pool volumes as the number of
clients you
        will backup in parallel.  This ensures that each client gets a
storage pool
        volume to dump data to on the server.

        Also, if you have multiple disks spread out the volumes over these
disks for
        better I/O performance.

        Hope this helps.

        Ramesh
        > -----Original Message-----
        > From: Masa, Ralf [mailto:masa.ralf AT COMBIBLOC DOT COM]
        > Sent: Thursday, September 02, 1999 9:11 AM
        > To: ADSM-L AT VM.MARIST DOT EDU
        > Subject: disk-pool performance
        >
        >
        > Hi ADSM-Folks,
        >
        > for performance reason: is there any recommendation for the size
of
        > disk-storagepool-volumes.
        > I mean, if I have one disk and one storage pool on this disk,
        > is it better
        > to have one large stg-pool volume or shoul I split the disk
        > into many small
        > stg-pool volumes ?
        > TIA
        >
        > R. Masa
        > masa.ralf AT combibloc DOT com
        >