UNSOLVED
SAN Migration Performance Evaluation
Does anyone have any useful methods for evaluating the performance requirements of a server with direct attached storage in preparation for migrating the server to SAN-based storage?
I continuously find myself at odds with developers and programmers that state they need a certain number of disks in a certain type of RAID set because their old server with direct attached storage had it and they don't want performance decreases when migrating to the SAN. I have begun a lengthy discourse with them regarding the performance characteristics of SAN based storage versus direct attached storage and am starting to convince them that designing LUN layouts in a SAN is not the same as in direct attached storage. The developers themselves have no idea what the performance characteristics of their application are, so it is up to me to evaluate the performance requirements and come up with a sensible design.
But the questions still remains in how best to evaluate the i/o requirements of a server when migrating it to the SAN. I am trying to find a way to translate native performance monitoring (Windows 2003 servers) to the i/o and throughput ratings of CLARiiON arrays.
Any help would be appreciated,
Aran
I continuously find myself at odds with developers and programmers that state they need a certain number of disks in a certain type of RAID set because their old server with direct attached storage had it and they don't want performance decreases when migrating to the SAN. I have begun a lengthy discourse with them regarding the performance characteristics of SAN based storage versus direct attached storage and am starting to convince them that designing LUN layouts in a SAN is not the same as in direct attached storage. The developers themselves have no idea what the performance characteristics of their application are, so it is up to me to evaluate the performance requirements and come up with a sensible design.
But the questions still remains in how best to evaluate the i/o requirements of a server when migrating it to the SAN. I am trying to find a way to translate native performance monitoring (Windows 2003 servers) to the i/o and throughput ratings of CLARiiON arrays.
Any help would be appreciated,
Aran
- Navin,
Agreed. Array cache and PowerPath load balancing (I always by full licenses) are in the design discussions I have with the developers. I was looking for more of a comparison matrix to use in evaluating servers for a SAN migration. I was looking for either the right combination of perfmon counters (it seems like there is an endless amount of combinations) or third party tools that I can use to evaluate a server's disk usage and ensure that I am providing adequate performance in a storage design.
Thanks,
Aran - Thanks Bowling, good article and recommendation on the SQLIO tool. That is a good idea to generate some baseline numbers for comparison design specs using those tools.
If I can at least get a baseline for various storage platforms that will help in evaluating design requirements.
And yes, simulating the application load is the trick as I have noticed that our developers tend to spend a lot of money on infrastructure they don't need then claim that the equivalent is required in the SAN because that is what the server had installed. One of my big challenges is determining how much i/o the application is actually generating so I can realistically scope the storage requirements at an application level and not just at a hardware comparison level.
Thanks
Aran - Aran,
One thing I've done in the past and continue to due, (time permitting), is perform various tests with SQLIO.
I generate a sample test and run them against local drives and then against SAN volumes of various configurations (RAID types, # Drives, Offset Settings, Allocation Size, etc.) and keep track of the tests on a spreadsheet.
The hard part is simulating the proper application load, sequential vs. random, io size, etc. for the application in question.
There have been times where we have been able to prove RAID5 performs better for sequential bulk loads verses RAID10, etc.
Also, you can reference the following MS artical for some disk counters to look at with expected values. The example is an Exhange Server but the same rules usually apply to most systems: http://technet.microsoft.com/en-us/library/aa997558.aspx
Hope this helps a little.
Navin-Swn_9
66 Posts
309
0
Posted July 31st, 2007 09:00
Windows you can use "perfmon" to count I/O on the disks. There are some tools available on Microsoft resource kit for the same.
You can record the readings and later on compare them with the Clariions.
If you will be using PP full on the windows box rest assured that you will get a better performance than a direct attach.
Thanks,
Navin.