Am Dienstag, 4. April 2006 16:46 schrieb Andreas Mohr:
Hi,
On Wed, Apr 05, 2006 at 12:36:06AM +1000, Con Kolivas wrote:
On Wednesday 05 April 2006 00:34, Andreas Mohr wrote:
And this all should work perfectly well with NON-soft-realtime scheduling, as clearly said before. Well, in theory, at least...
Andi just out of interest, how does normal scheduling on current ck (2.6.16-ck3) perform with this?
Hmm, difficult :-\ I don't have any game candidate here, and frankly I don't even have a current Wine install...
No need for a full-blown game, I use this great free 5.9MB demo to test audio on Wine: http://www.pouet.net/prod.php?which=18359
Unbearable with regular Wine (kernel 2.6.13/ amd64, gentoo r5 patchset - can't test with later kernels, as there's a problem with the SCSI-subsystem as of 2.6.14), just dandy with Mikes patch and realtime-lsm...
I'd think it's much easier for someone with such a Wine testcase to use a new kernel:
- locate the .config file of the currently installed kernel package
- get 2.6.16, patch to 2.6.16-ck3
- run "make oldconfig" to answer all differences from current kernel to 2.6.16-ck3 only
- make bzImage modules modules_install
- (update bootloader)
About my sys call preemption latency rantings: It would be very useful to experiment with various CONFIG_PREEMPT settings here. And if some anomalies turn up, then it might be useful to add Ingo Molnar's latency tracing patch to a kernel and debug it further.
Stuff like that really shouldn't happen given that many people (among those also Windows core developers) say that our scheduling and thread creation performance usually beats Windows XP hands down.
Andreas