Need performance impact for coherency traffic performance counters

[John_Rudelic]

All, I can use Vtune to generate counts for the coherency traffic events but there is no documentation that tells me the impact/penalty assiciated with each event counter. 

There are come offcore event counters that get to billions of operations while there are other offcore event counters that get to millions of operations. 

There is no way to tell which event counters impact cohernece traffic/performance and how much each event counter impacts coherence traffic/performance.

I am specifically interested in understanding how coherency traffic impacts the performance of my driver/app as I move it around in a NUMA system.

Thanks 

[Peter_W_Intel]

It says, "...A minimum latency of 32 cycles should give a reasonable distribution for all the offcore sources however." when L3 MISS (local/remote DRAM go to S/E) in B.2.3.2 of this article

Does it help?

Regards, Peter