Re: [linux-audio-dev] Performance and Elegance? (Was: High Cost of IPC)

New Message Reply About this list Date view Thread view Subject view Author view Other groups

Subject: Re: [linux-audio-dev] Performance and Elegance? (Was: High Cost of IPC)
From: John Regehr (regehr_AT_cs.utah.edu)
Date: Fri May 18 2001 - 00:19:32 EEST


Peter said:

> What's a simple way to measure the working set of a process or thread.
> I'd like to check it on my project.

I took the easy way out and used a synthetic application: threads
repeatedly walk private arrays, and the working set size is defined to
be the array size. In real life it's not so easy! One thing to do
would be to borrow or write some instrumentation code that uses the P6
MSRs - these include rdtsc as well as counts of misses to various
caches, number of pipeline stalls, etc. - by checking appropriate MSRs
at the right times you could put together a pretty accurate picture of
what's going on in the caches. Not sure how to distinguish between
different kinds of misses. Maybe run a single thread in isolation (is
this possible?) and then together with others. Any additional misses
experienced by this thread in a multithreaded environment can be assumed
to be caused by context switches. Documentation for the MSRs
(model-specific registers) can be found at the Intel site.

John


New Message Reply About this list Date view Thread view Subject view Author view Other groups

This archive was generated by hypermail 2b28 : Fri May 18 2001 - 00:48:21 EEST