Thread wake-up latency under Windows Server R2

[moshiko]

Hi,

I have an standard producer-consumer application which uses signaling, so that the producer signals the consumer to wakeup whenever it has a new data for processing.
For performance analysis, the application measures the time (in usec) that it takes to wake-up the consumer, that is the time from the first signal sent to consumer while it was sleeping, till it wakes up.

I have run this application in 2 environments: both with Windows Server 2008 R2 (Service pack 1), 32GB RAM, same .NET framework installed.
Environment (A) has Xeon E5450, while Environment (B) has Xeon L5640.
Besides that the settings are pretty much similar.
Application is defined to run in "realtime" priority level.

The latency measurements show ~100usec on (A) while much smaller latency around ~10usec on (B).

I am trying to figure out the cause for this difference, my questions are:
1. What is the way to analyze the difference? Can Windows Performance Monitor be helpful ?

2. I suspect that the difference comes from (A) having only 4 cores, while (B) having 8 cores and multi-thread support up to 12 hw threads. So, in (B) the OS have more resource to avoid context-in and out of consumer while it is sleeping. Is there any way to restrict OS to put my application in pre-defined core, and not to use that core for any other threads ? (affinity gives only the first part...)

Any answer/idea would be appreciated.

Thank you,
-Moshe

[Arjay]

Are there other differences between a) and b) like a difference in what applications are running?

[moshiko]

No other difference between A and B. Both stations basically run only my producer-consumer app. This is why I dont see any reason for this latency