I/O subsystem requirements for the tempdb database

re: I/O subsystem requirements for the tempdb database

Jake Marx — Sat, 16 Sep 2006 02:07:39 GMT

Travis - have you implemented the ramdrive yet? Any thoughts or observations to share on it?

Thanks!

re: I/O subsystem requirements for the tempdb database

Travis — Mon, 14 Aug 2006 19:23:32 GMT

We are getting a TMS ramdrive in sometime this week. I know that our app doesn't have any disk queing to speak of, so it will be interesting to see if we get a performance increase or if we just move the load to somewhere else in the server.

Either way, I will check back in and share the results of the benchmarking.

re: I/O subsystem requirements for the tempdb database

Euan Garden — Sat, 08 Jul 2006 08:41:40 GMT

Jake,
As a general rule Microsoft does not certifiy 3rd party storage products. RamSan may well work but it will not have been tested by the SQL Team and hence its impossible to its fine to use it.

If under your testing it works just fine and the vendor is willing to provide you with support/integretity assurances then its your call.

re: I/O subsystem requirements for the tempdb database

Jake Marx — Thu, 29 Jun 2006 23:31:13 GMT

Hi Euan,

The KB article states "an implementation such as a RAM disk may only be used as the location of the tempdb database and cannot be used for any other database" due to the fact that it's not "durable media". However, there are several products out there (such as the RamSan from Texas Memory Systems) that claim to work well with SQL Server: http://www.texmemsys.com/files/f000174.pdf

Since products like this have "hot" backups and use "chipkill" technology to detect and recover from errors, would they be considered to be "durable" and thus suitable for database files (other than tempdb)?

We're considering moving to SSD for one of our very busy databases (having issues on an EMC Celerra with high I/O & log wait). Our tempdb is OK based on all the performance monitoring we do, but the LUN the data file is on is performing poorly. I'm certain an SSD-based disk will help with I/O issues, but is doing this against MS's recommendations?

Thanks,
Jake

Interesting Finds: June 29, 2006 AM edition

Jason Haley — Thu, 29 Jun 2006 18:06:19 GMT