June 19, 2012 at 1:19 pm
chris-320654 (6/19/2012)
I re-ran my tests with hyperthreading disabled and there was no benefit at all. If anything times were very slightly slower for the single threaded powershell test, and noticeably slower for the application itself. CXPACKET remained the top wait stat at around 73% and LATCH_EX at 26%.I know I can eliminate CXPACKET by increasing cost threshold for parallelism but this does affect other parts of the application which would run slower as a result. Is there anything else I can do to reduce this type of wait?
1) glad to hear things are running better now!
2) I am not at all surprised that hyperthreading didn't make your OLTP-type app run faster.
3) "sql server 2005 waits and queues" -that white paper is the Bible when it comes to that topic. It is for SQL 2005 but still VERY applicable. In it you will find track_waitstats_2005 and get_waitstats_2005, sprocs to do time-differential wait stats analysis. http://technet.microsoft.com/en-us/library/cc966413.aspx
4) Speaking of which, the ratio of wait stats on your box isn't particularly meaningful. What matters is how MANY waits of a particular type and how long they lasted. Having said that, CXPACKET waits often indicate suboptimal indexing in an OLTP scenario. The LATCH_EX could be BAD if the waits are on tempdb PFS/SGAM pages, indicating temp object allocation issues. Both of those have pretty simple fixes thankfully. Some other latch issues can be more difficult to decipher/resolve.
5) Feel free to drop me a line if you decide need some professional help. If not me, there are quite a number of excellent resources here on SSC.com!
Best,
Kevin G. Boles
SQL Server Consultant
SQL MVP 2007-2012
TheSQLGuru on googles mail service
Viewing post 31 (of 30 total)
You must be logged in to reply to this topic. Login to reply