Subject: Re: [linux-audio-dev] Linux and clocks
From: Erik Walthinsen (omega_AT_temple-baptist.com)
Date: Fri Mar 22 2002 - 00:06:32 EET
On Thu, 2002-03-21 at 13:47, Martijn Sipkema wrote:
> Is there something that can be easily done about the bad
> performance of clock_nanosleep at long sleep times?
Yes, there are things called 'firm timers' that use the IO-APIC to
(transparently) give applications extremely high resolution on longer
sleeps. I've cc:'d the guy who's been working on the patch most
recently, maybe he's got a draft of his paper he's willing to post.
The quick summary of the paper is that with firm timers he got the
jitter of the usleep() used in MPlayer to almost zero, dead on at 33ms
every time. Without the patch, the wakeup times cluster relatively
cleaning at either 30 or 40ms, as expected with HZ of 100.
(for reference, short usleeps are done as a busyloop, which explains the
high accuracy)
Erik Walthinsen <omega_AT_temple-baptist.com> - System Administrator
__
/ \ GStreamer - The only way to stream!
| | M E G A ***** http://gstreamer.net/ *****
_\ /_
This archive was generated by hypermail 2b28 : Thu Mar 21 2002 - 23:55:34 EET