Collocation also helps because you are able to run parallel restores more
effecitvely since there will be no media contention. i.e. restore of
filespace A will not request the same tape as restore of filespace B.
Therefore the overall restore throughput is much better.

As far as the restore of small files. This appears to be a major issue with
Windows NT although no operating systems are free from this problem either.
I believe that a significant bottleneck lies within the NTFS filesystem.
Significant performances improvements can be obtained by using faster disk
and processor on the client.
Earlier this week we restored a 200Gb NTFS drive (a mixture of small 2kf
files and 200Mb large files) in just over 10hrs. For the purpose of the
restore we used a faster FC-AL array as opposed to SCSI-2 and increased from
2 x 200Mhz Pentium Pro to 2  x 500Mhz Xeons. We also raised the
TCPWindowsize to 1024 on the NT Client.

The introduction of the backup set (tivoli 3.7) should help in the cases of
small files, but I haven't done any real tests yet to see what sort of
performance gain is had.

Nathan









        -----Original Message-----
        From:   Jason Morgan - I.T. [SMTP:Jason.Morgan AT BLAENAU-GWENT.GOV DOT 
UK]
        Sent:   Monday, November 22, 1999 10:41 AM
        To:     ADSM-L AT VM.MARIST DOT EDU
        Subject:        Re: Speeding up restores

        Hi Chris,

        Collocation will help because files are grouped together by Client
Node or
        Filespace therby requiring fewer media mounts on restore processing.
On the
        down side more media mounts are required to collocate files and also
more
        mounts are required during reclamation and storage pool processing.

        Cheers Jason


        > -----Original Message-----
        > From: Jordan, Chris (ELS) [SMTP:c.jordan AT ELSEVIER.CO DOT UK]
        > Sent: 22 November 1999 16:11
        > To:   ADSM-L AT VM.MARIST DOT EDU
        > Subject:      Speeding up restores
        >
        > Are there any clever ideas for speeding up restores??
        >
        > We have 5 or 6 servers with millions of small files. It is these
that are
        > causing the thought processes. It is expected in the event of a
disaster
        > that these may take 5 or 6 days to restore. The slow restore is
due to
        > both
        > the small files, and the number of tapes that will be required as
the data
        > to be restored will be spread over many different tapes written
over the
        > last 3 months or so.
        >
        >
        > Can we take an export once a week. Use this in the event of a
disaster to
        > import a known state, restore using this, and then restore the
remaining
        > changes using the normal backups?
        >
        > Will collocation help enough to make it worth while?
        >
        > Can we produce a tape that can be used to restore all or a
majority of the
        > system?
        >
        > Any other ideas??
        >
        >
        > Cheers, Chris