Messages in this thread | | | Date | Tue, 30 Nov 2004 15:39:58 -0000 (WET) | Subject | Re: Real-Time Preemption, -RT-2.6.10-rc2-mm3-V0.7.31-7 | From | "Rui Nuno Capela" <> |
| |
> Ingo Molnar > > * Rui Nuno Capela wrote: > >> xruntrace1-2.6.10-rc2-mm3-RT-V0.7.31-13-*.trc.gz >> - the captured traces, as dumped with xruntrace1_watch.sh >> script. > >> Each trace shows only the first XRUN occurrence on a distinct jackd >> session. Every other trace were triggered and captured after >> restarting jackd. > > interesting - the trace shows only a latency of 20-40 usecs. Perhaps the > tracing should be done over two periods? > > Could you try to hack alsa_driver.c to do the gettimeofday(1,1) call for > only every second case. Something like this ought to be enough: > > { > static count = 0; > > if (!(count++ & 1)) > gettimeofday(0,1); > } > > (only one thread accesses this particular variable so there are no > threading issues.) > > with this variant there's a 50% chance that we get the trace of the last > 2 poll() instances. (and there's a 50% chance that we get only 1 period > covered.) > > in any case, the scripts & approach seems to be almost there, i hope we > only need the above change to see the true source of the xruns. >
Done that.
New XRUN traces are attached, while running RT-V0.7.31-15 now. However, I don't seem to get any notorious difference on the results, since previous ones. All latencies traced ca. 26-27 usecs.
Bye. -- rncbc aka Rui Nuno Capela rncbc@rncbc.org[unhandled content-type:application/x-gzip-compressed][unhandled content-type:application/x-gzip-compressed][unhandled content-type:application/x-gzip-compressed][unhandled content-type:application/x-gzip-compressed][unhandled content-type:application/x-gzip-compressed] | |