On Tue, Apr 26, 2011 at 05:25:38PM -0700, John Gilmore wrote: > Also, Gnash still burns a lot of CPU time (but not 100%!) after > showing a youtube video. It's better...for example this page: > > http://www.villagetelco.org/2011/02/in-a-village-telco-minute/ > > seems to burn about 11-15% of one CPU when doing nothing (just waiting > for someone to click on the big centered Play button to start the > video coming in from youtube). Strace shows lots of poll() calls with > timeouts of 0 ms or 9 ms, plus dozens of gettimeofday() calls, an > ioctl(,FIONREAD) and a read() returning EAGAIN, continuously.
An heartbeat is dispatched every 10ms, by design. Would be interesting to inspect further for the 0ms timedout poll. The gettimeofday are due to the VirtualClock being checked at least on every heartbeat but probably more than that. They might be reduced to one-per-heartbeat by pretending time doesn't flow between them. Another time checker is NetStream, btw. Would be interesting to compare runs with and w/out videos to see how gettimeofday calls change. The ioctl and read I'm not sure about. Any chance to get a call stack ? --strk; () Free GIS & Flash consultant/developer /\ http://strk.keybit.net/services.html _______________________________________________ Gnash-dev mailing list [email protected] https://lists.gnu.org/mailman/listinfo/gnash-dev
