Please start any new threads on our new
site at https://forums.sqlteam.com. We've got lots of great SQL Server
experts to answer whatever question you can come up with.
| Author |
Topic |
|
igormarchenko
Starting Member
5 Posts |
Posted - 2009-12-18 : 18:51:15
|
| Hello! I realize this give questions might be too generic, but hopefully somebody will point me into right direction. We in the process if migrating our SQL Server 2005 database Active/Active cluster to 2008. We have setup and configured SQL Server 2008 cluster and I started testing/comparing performance between old and new clusters. Database in question has been copied over to 2008, upgraded, statistic updated and compatibility level changed to 100. I noticed that very same stored procedure generated different plans in 2005 and 2008. It ran slower on 2008. Did anyone has similar experience when stored procedure ran slower after upgrade to 2008? I have also noticed that very same query that generates identical plan ran slower on 2008:select * from <tablename> --tested tables with 300K and 500K rows SQL Server 2008 cluster has LUN shared by Data, Log, TempDB drives (not exactly the best practice; has been done to cut costs). Also, SQL Server 2008 -64 bit (SP1 CU2) has 32GB of RAM and 8 CPU ,whereas SQL Server 2005 -64-bit (SP2) has 16 CPU with 64GB. Parallel execution is disabled on both servers. 2008 server has 'lock pages in memory' configured I compared performance of repetitive calls when data pages are in cache and confirmed that number of physical reads and read ahead reads were 0 indicating that data pages were read from memory. Number of logical reads was pretty much identical but overall duration on 2008 was somewhere between 25%-40% slower. I assume that since data pages are in memory the only difference is RAM characteristics. The fact that LUN(s) wasn't configured based on best practices on 2008 server shouldn't be affecting performance when data pages are in cache, correct? Are there any hardware cache, HBA (other?) settings that worth looking at? Any advice on this matter is appreciated.Igor |
|
|
|
|
|