> > > PEBS is important in our NUMAtop design to measure > memory access > > latency for application. We will enhance kcpc to > support PEBS and > > uncore events. If we can talk with CPC engineers > and get their help, > > it will be great. > > Yes, I think that's essential. I'll put you guys in > contact > with each other.
Actually, I am the one who is prototyping a PEBS sampling framework. Right now, the prototype works only on the Intel core platform and only supports cpu-binding. I am going to add the Nehalem support and thread-binding in the near future. However, please keep in mind that it is going to take some time to get this framework available in OpenSolaris as it involves multiple components that require stable design and implementation. If you need further information on the PEBS work, please feel free to contact me at krishnendu.sadhuk...@sun.com. -Krish > >> I'm sure sampling is probably what you need to get > the job done > >> properly. However, don't forget about the DTrace > cpc provider > >> as it may well be of use to you here. > >> > > We have considered CPC provider in our design to > help user locate what > > function caused most of remote memory access, but > this feature will > > not be in current implementation. Both NUMAtop and > CPC provider need > > access performance counters, they can not program > performance counter > > simultaneously. When customer identifys that his > application may be > > impacted by remote memory access through NUMAtop. > If he want to know > > the top functions which leads to remote memory > access, NUMAtop can > > enter into CPC provider mode for sampling and > provides such > > information. > > OK. If we could implement a sampling framework it > would be a > natural and highly useful extension to provide a > sampling > provider for DTrace. We've spoken about that > internally but I'm > not 100% sure what it would look like at the minute. > All food > for thought though. > > Jon. > _______________________________________________ > perf-discuss mailing list > perf-discuss@opensolaris.org -- This message posted from opensolaris.org _______________________________________________ perf-discuss mailing list perf-discuss@opensolaris.org
