ADSM-L

AW: TSM Performance - Is it my backuppool

2003-03-07 04:54:53
Subject: AW: TSM Performance - Is it my backuppool
From: Salak Juraj <j.salak AT ASAMER DOT AT>
To: ADSM-L AT VM.MARIST DOT EDU
Date: Fri, 7 Mar 2003 10:55:20 +0100
Hi!

due to concepts used Raid5 is the slowest solution available for sequential
writing.
Raid5 reads are always much faster than Raid5-5 writes, 
simply because reads require less I/O´s on disks than writes do.

However, 1,5MB/sec means you have got either very slow controller 
or your controller´s / Raid configurations need optimising.

Check for cache settings on both your controller and your disks,
they are probably set to write-through.
Change both to write-back and give it a test.
This is a no-problem setting if you have battery on the controller,
otherwise think twice if you want use this setting.

Another setting to check is Stripe-size 
(basically the amount data which is physically written
on raid at once). 
For sequential writing you may want to set it as large as possible.
Do not do this for writing of small files or for database applications.

Yet another point is amount of physical disks in your raid5.
Raid5 wit 3disks is relatively slow, 
with overall performance raising significantly when using 5 to 8 disks.


Raid5 is quite preferment when used for parallel random reads 
and fairly good when using for parallel random mixed R/W operations
like more storage pools on one Raid 5 used for storing files of mixed sizes.

Raid 1 is about as performant as JBOD for writing and 
about as twice as performant when used for reads with
broad spread depending on application - from same speed as JBOD
up to n-times faster, while n is seldom larger than 2.

Raid10 depends on manufacturer and application 
- I saw performances outperforming JBOD n times,
where n is the amount of disks in raid-10,
But I also saw Raid10 with performance slower than JBOD.


Regards
Juraj Salak



-----Ursprüngliche Nachricht-----
Von: Farren Minns [mailto:fminns AT WILEY.CO DOT UK]
Gesendet: Donnerstag, 06. März 2003 16:49
An: ADSM-L AT VM.MARIST DOT EDU
Betreff: TSM Performance - Is it my backuppool


OK, I have just done another test and here are my findings.

TSM 4.2.2.12 running on a Solaris E250, 400Mhz, 1GB, 50GB backuppool on
RAID5 array, 8500MB DB and LOG on separate internal disks (TSM mirrored).

I just backed up a 100Mb file via a client on the same machine as the
server. Data transfer rate 1.5MB/s. Terrible. File was on the same RAID 5
array as the backuppool. I then restored that very same file to the very
same location and got a transfer speed of no less than 51MB/s. Much better!

So, could the problem lie with my backuppool for some reason, and if so,
how would I know. We do have caching turned on if that makes any
difference.

Any ideas on this one?

Thanks to all

Farren Minns - John Wiley & Sons Ltd

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