Any advice for using IOMeter to baseline an EVA8000?

[GilchristT]

Hi

I'm trying to use IOmeter to provide a baseline set of statistics for a new EVA8000. I want to check things like VRAID5 performance vs VRAID1, using 8 disks in a disk group vs. 32 disks, etc.

The problem I'm seeing is that the figures are initially very high but tail off. I'm guessing this is down to cacheing but it seems to take many hours to get to a stable value. Is the cache that good?!

All the boxes connected are Wintel running Windows 2003. The system is a sandbox at the minute and I can make any changes suggested but I lose that option in about a week.

Any suggestions or voices of experience welcome.

thanks

tommy

[florence mathon lermusi]

Hi Tommy,

the first thing is that before you start your tests you must have done some initial writes on your disks, because the eva keeps tracks of writes and if your disks have never been accessed to write data on it the eva will send an automatic answer without checking the data on disks until you are writing something on it during your tests etc...

[Uwe Zessin]

The latest "best practices guide" for the EVA says that this is a myth.

[Peter Hoogenboom Sijtsm]

Tommy,

Caching can do a lot but it all depends on the profile(s) created in IOmeter. You really have to determine what you want to test up front, know what the properties of your appliactions are and dig into the iometer manual. I believe IOmeter is advised to run for 24 hours for creating a baseline. Don't forget to create the IOmeter file with the size of your database.

If you mean sandbox as in SAP with SQL you can better use SQLIO for performance measurements. There where some transactions IOmeter gives false measurement for and SQLIO doesn't.
Don't forget to tune your windows systems before testing the systems and putting data on your disks (change the diskoffset with diskpar.exe (not diskpart!!) and for SAP with SQL you can best format them with an allocation unit size of 8192 bytes).

I hope this helps.

Good luck,

Peter.