On Tue, 6 Nov 2007, Josh Carroll wrote:
That's expected due to the fuzzy rounding of 128 / 10, etc. Can you set
slice_min and slice both equal to 7 and see if the numbers come out
better than without the patch but with a slice value of 7? Basically I'm
trying to isolate the effects of the different slice handling in this
patch from the other changes.
Here are the results. Looks like setting slice_min and slice to 7
degraded performance for all work loads. Should I try any other
settings for slice and slice_min?
Josh, I had an interesting thought today. What if the reason 4BSD is
faster is because it distributes load more evenly across all packages
because it distributes randomly? ULE distributed across cores evenly but
not packages. Can you try the attached patch? This also turns the
default slice size down but does not contain the other context switch
performance improvements.
Thanks,
Jeff
ffmpeg: 1:39.91
sysbench (4, 8, 12, 16 threads respectively):
2013.78
2157.6
2175.91
2184.16
And buildworld time: 14m7.517s
Thanks!
Josh
_______________________________________________
freebsd-performance@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-performance
To unsubscribe, send any mail to "[EMAIL PROTECTED]"
Index: sched_ule.c
===================================================================
RCS file: /home/ncvs/src/sys/kern/sched_ule.c,v
retrieving revision 1.216
diff -u -r1.216 sched_ule.c
--- sched_ule.c 23 Oct 2007 00:52:24 -0000 1.216
+++ sched_ule.c 9 Nov 2007 23:01:21 -0000
@@ -1105,6 +1105,9 @@
* Look for an idle group.
*/
CTR1(KTR_ULE, "tdq_idle %X", tdq_idle);
+ cpu = ffs(tdq_idle & 0x5a5a5a5a);
+ if (cpu)
+ return (--cpu);
cpu = ffs(tdq_idle);
if (cpu)
return (--cpu);
@@ -1324,7 +1327,7 @@
* in case which sched_clock() called before sched_initticks().
*/
realstathz = hz;
- sched_slice = (realstathz/10); /* ~100ms */
+ sched_slice = (realstathz/15);
tickincr = 1 << SCHED_TICK_SHIFT;
/* Add thread0's load since it's running. */
_______________________________________________
freebsd-performance@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-performance
To unsubscribe, send any mail to "[EMAIL PROTECTED]"
