I know I probably shouldn’t be replying to this old thread, but I’m crossing my fingers that Jonathan Kehayias will see activity on this old thread and respond…
I’m using Quest Benchmark Factory to run the TPC-E workload on some SQL instances, and comparing results. I’ve got 1 server built to our company’s standards (standard hardware, disk, etc.). It’s Windows 2008 R2. I’ve installed SQL 2008, SQL 2008 R2, and SQL 2012. One by one, I’ve run the TPC-E workload against each SQL instance. I’m seeing much better throughput/response time on the SQL 2008 and R2 instances than what I’m seeing on SQL 2012. When I dig into it, the thing that stands out to me is that I’m seeing much higher numbers of tasks in the runnable queue (looking at sys.dm_os_schedulers, and sys.dm_exec_requests). For example, when running against SQL 2012 I see the total number of runnable tasks up around 40-70, yet the CPU is around 40%.
Also, in looking at sys.dm_os_schedulers, the 2008 instances appear to have a much better distribution across the schedulers. I see more “active” schedulers (is_idle column). On 2012, I see fewer schedulers active. It seems a small handful are hit pretty hard while the majority are inactive.
This is a 16 core machine with HT enabled, giving me 32 procs total. I’m running standard edition at the OS level, as well as SQL. My initial thought was that SQL 2012 Standard wouldn’t use all 32 procs, but it seems to be. (I know SQL 2012 Standard is limited to 16 *physical* cores)
I’m hoping someone may have some insight. I very clearly have tasks sitting in the runnable queue, yet my CPU is < 40%. Not sure where to look next. Any help would be appreciated.