On Thu, Oct 25, 2018 at 10:59:36AM +0300, Alexey Budankov wrote: > Hi, > > On 15.10.2018 13:17, Jiri Olsa wrote: > > On Mon, Oct 15, 2018 at 09:26:09AM +0300, Alexey Budankov wrote: > >> > >> Currently in record mode the tool implements trace writing serially. > >> The algorithm loops over mapped per-cpu data buffers and stores > >> ready data chunks into a trace file using write() system call. > >> > >> At some circumstances the kernel may lack free space in a buffer > >> because the other buffer's half is not yet written to disk due to > >> some other buffer's data writing by the tool at the moment. > >> > >> Thus serial trace writing implementation may cause the kernel > >> to loose profiling data and that is what observed when profiling > >> highly parallel CPU bound workloads on machines with big number > >> of cores. > >> > >> Experiment with profiling matrix multiplication code executing 128 > >> threads on Intel Xeon Phi (KNM) with 272 cores, like below, > >> demonstrates data loss metrics value of 98%: > >> > >> /usr/bin/time perf record -o /tmp/perf-ser.data -a -N -B -T -R -g \ > >> --call-graph dwarf,1024 --user-regs=IP,SP,BP --switch-events \ > >> -e > >> cycles,instructions,ref-cycles,software/period=1,name=cs,config=0x3/Duk -- > >> \ > >> matrix.gcc > > > > I ran above on 24 cpu server and could not see the gain, > > but I guess I'd need much bigger server to see that > > > > anyway, the code is now nicely separated, and given the > > advertised results below I have no objections > > > > Reviewed-by: Jiri Olsa <[email protected]> > > Is the plan Jiri mentioned earlier to have it as a stand alone patch kit > or upstream the changes into mainline?
I haven't heard from Arnaldo yet, but I'd like to have this merged in jirka
