ADSM-L

Re: Large server restore time?

2002-01-16 09:44:39
Subject: Re: Large server restore time?
From: "Seay, Paul" <seay_pd AT NAPTHEON DOT COM>
Date: Wed, 16 Jan 2002 00:49:26 -0500
You will be lucky to get 30gb/hr.  Typically about 7 to 8 MB/sec is all you
will ever see, maybe 10 if your files are larger over a 100mb interface.
Also, not all interface cards and  Ethernet card drivers are created equal.
Many are 100mb connectivity, but really cannot perform at continuous
throughput of more than 60mb.  There are many ways to help which various
responders have suggested.

Windows allocation is slow at best and single thread to a file share or
drive letter which ever applies based on some documents that I have
reviewed.  One suggestion was to make sure you create multiple file spaces,
but also you need to use collocate by filespace to make sure the data is
spread across the tapes so that parallel restore commands could be issued
from the client.

The other pieces to this equation are what kind of disk subsystem is the
windows server going to be attached to.  If it is something like an IBM ESS,
HDS 9900, or EMC 8730 with FC connections, then it will perform pretty well.
The other piece is how many tape drives did you put on the SCSI channels
each.  I prescribe to 1 adapter to 1 drive on these high end drives if you
really have the network bandwidth.  But there is no way you are doing that.
I bet you are 1 adapter to 3 to 5 drives.  This causes the drives to collide
on the SCSI bus and potentially dramatically slow down if the are active.

However, all said and done, you have many issues to resolve for restoring a
server of this magnitude, from experience.  Be prepared to roll out your
wallet.  Gigabit connectivity, parallel pathways, CISCO switch upgrades,
converting to SAN tape, etc are in the offing.  This has nothing to do with
TSM it is just physics.

<Prev in Thread] Current Thread [Next in Thread>