Sql-server – DBCC sqlperf(waitstats) contains signal wait time longer than last time waitstats was cleared

Thank you for the detailed explanation of your problem (one of the best laid out questions actually).

WRITELOG is a very common type of wait, so don't worry about it. Looking at the SOS_SCHEDULER_YIELD indicate CPU pressure and also the CXPACKET, it is possible that there must be some missing indexes and you may be retrieving lot of data from the queries for an OLTP system. I suggest you to look at the Missing Indexes DMV and see if there are any indexes (almost sure there will be more than few) that are in the questionable procs.

http://sqlfool.com/2009/04/a-look-at-missing-indexes/

http://troubleshootingsql.com/2009/12/30/how-to-find-out-the-missing-indexes-on-a-sql-server-2008-or-2005-instance-along-with-the-create-index-commands/

Look for Jonathan Kehayias's post on sqlblog.com on this too.

Also, take a look at Parameter sniffing.

http://sommarskog.se/query-plan-mysteries.html

http://pratchev.blogspot.com/2007/08/parameter-sniffing.html

It's NOT a compete answer for your needs but a good starting point. Let us know if you need more details.

So I resolved this, turns out that power management features were enabled on our SQL server that were scaling the CPU frequency up and down, but not fast enough to keep up with the small demand and introduced the SOS_Scheduler_Yield wait. After changing it to run always in high performance the issue went away and now the waits are more normal (LatchIO type stuff).