In my case I have my file deviceclass for one particulair node on one
big raid 5 array. ( 8x 250GB sata). This node has a lot of small files
(emails).
A restore with just one session going only gave me a throughput of
around 1 - 2 MB per second.
A restore with 8 sessions gave me a throughput of around 10 MB per
second, this is my network bottleneck (100 mbits)
With online db2 backups 2 sessions was faster then 1 sessions to the
same array although I can't remember the exact numbers.

So my general rule of the thumb is "more sessions is better" but you
want to test of course what is better for you and let us know. You might
want to search the list-archive because this discussion has been done
before.

cheers

Otto Schakenbos
System Administrator

TEL: +49-7151/502 8468
FAX: +49-7151/502 8489
MOBILE: +49-172/7102715
E-MAIL: oschakenbos AT teleflex DOT com

Corporate IT Europe
Teleflex Holding GmbH
Fronackerstrasse 33-35
71332 Waiblingen
GERMANY




Allen S. Rout wrote:

==> On Tue, 6 Dec 2005 11:58:01 -0500, William Boyer <bjdboyer AT COMCAST DOT 
NET> said:

If you collocate you run the risk of not being able to use the multi-session
restore capability of the client. As for the n-servers backing up you will
have N-files open...what if you're running a RESOURCEUTILIZATION higher than
2 on the client? You can have more that N files open if the clients have
more than 1 data backup session running.

I agree about possibly having more than N files open for N clients; a high
mountlimit is important.

My guess would be that multisession capability won't be an important factor in
restore speeds once you have all the data on (de facto) random access media.
I'd guess we'd be back to network pipe or write speed of the recieving disk or
filesystem.

My experience on raids of SSA has been that trying to read multiple large
files simultaneously hurts my performance quite a bit; from ~27 MB/s sustained
read for a single thread, down to ~21 (total) if I try to read two, and ~18
(total) if I try to read three.  This will of course be ludicrously
disk-subsystem-dependant; but it's at least one unambiguous case in which the
idioms of tape optimization led me down a bad path when I applied them to
disks.


- Allen S. Rout