ADSM-L

Re: [ADSM-L] Performance and migration: AIX vs Linux

2009-10-23 09:31:59
Subject: Re: [ADSM-L] Performance and migration: AIX vs Linux
From: Jerry Michalak <jerry_gjm AT YAHOO DOT COM>
To: ADSM-L AT VM.MARIST DOT EDU
Date: Fri, 23 Oct 2009 06:29:59 -0700
I had a similar setup about 2 years ago and ran into some throughput problems 
also.

First, you need to measure the speed of tape output when doing migrations, 
disk-to-tape, and some sort of tape-to-tape processes. I found with my setup 
that I was getting faster transfer speed when doing tape-to-tape !!!

Further investigation pointed to the Clarion device. We where sharing it with 
other applications and found that even though we had many spindles for TSM use 
that we were restricted by the fact that the Clarion system only had 2 I/O 
ports to the SAN running in a failover mode. The EMC people were called and 
"we" were driving the 2 ports at over 80% each so if we ever had a failover we 
were going to be in big trouble. By over using the ports the software within 
the Clarion was slowing our disk I/O down to something slower than the tape 
drives !!!

We went to local disks instead of SAN disks for all of the TSM servers after 
that.


 Jerry Michalak
jerry_gjm AT yahoo DOT com




________________________________
From: "Dury, John C." <JDury AT DUQLIGHT DOT COM>
To: ADSM-L AT VM.MARIST DOT EDU
Sent: Thu, October 22, 2009 5:13:23 PM
Subject: [ADSM-L] Performance and migration: AIX vs Linux

We are currently running TSM server v5530 under AIX. The AIX server has a 
mixture of different speed (266mhz and 133mhz, both 64bit) PCI-X slots. With 4 
4g HBAs. Our system is connected to a Clariion CX3-80 where the TSM DB and 
Recovery Log and Disk Storage pools live. The disk parts of TSM have 2 4G HBAs 
and powerpath installed and configured for load balancing across both. We have 
2 STK SL500 libraries, both with 4 LTO4 drives. One is local and the other is 
connected via a 2 2G fiber paths to our remote site.  The remote SL500 is 
defined as a copy storage pool. All of the local tape traffic is on one HBA and 
the remote is on another. Disk and tape do not have any traffic on the same 
HBAs ever.   I used a script I found in past emails and the best performance we 
can get on the local SL500 is around 30MBs. The best performance we can get on 
the remote library is about 10MBs.  We recently redid the AIX box and rezoned 
so the disks would be using the
 fastest PCI-X slots and the tape would be using the slower ones to try and 
improve disk performance. This did help some as expiration was running for 
about 8 hours and now it runs in about 4. Our TSM DB is 76800 MB and is 75% 
full right now. I've looked through the latest TSM performance and tuning 
manual and tweaked everything as per recommendations for our setup.

We are considering migrating our AIX server to an Intel based Linux box that 
has a mixture of 8x and 4x PCI Express slots in hopes of maximizing performance 
on both disk and tape and also saving money as AIX boxes are considerably more 
expensive.

So here are my questions:

1.        How hard is it to move from an AIX TSM server box to a Linux TSM 
server? I'm hoping it's as easy as building the new box (tape drive,stg pool 
etc) and then restoring the DB and tweaking the new config. I know there is 
more to it than that but without researching it yet, that seems like a logical 
high level overview.

2.        Will there be much of a performance difference between and AIX based 
TSM server and a Linux based TSM server?

3.       Going from mixed speeds slots of PCi-X to PCI Express 8x and 4x slots  
should be a significant improvement correct?

I know there are a lot of factors here but we are concerned we aren't getting 
the best performance for our existing hardware. 30MBs (not Mbps) for a LTO4 
drives seems pretty slow and with 4 4x PCI-E and 4 8x PCI-E slots, I can 
balance out the I/O across cards and slots  much better than I can now.

Comments and criticisms? Linux vs AIX?
Thanks for any insight!