For a given hour, my signal waits were 20% yet my OS CPU utilization during that time never went over 25%. Can someone explain why my signal waits were so high?
My vendor has a best in class scoring system that expects us to be at 10% or less signal waits, or we get dinged. How can I go about fixing this (without adding additional CPUs)?
- We do not have more than 8 CPUs per NUMA node, so Trace Flag 8048 does not apply.
- The largest instance waits are
cost threshold for parallelismis set to 50. Should I raise it? To what?
- This is a physical machine (not a VM), dedicated to SQL Server.
- I am using a monitoring tool to identify the most frequently run queries and procedures. Do I want to look at high CPU, high I/O, or high duration? Normally our app is I/O intensive so I tune high I/O. But since the issue is signal waits, do I need to look at high CPU?
- I was hoping to avoid Max Vernon's recommendation to lower
MAXDOPto 4, because the app does some warehouse style queries that need the extra threads.