Re: [rfc][patch] improved interactive starvation patch against 2.6.16

[Willy Tarreau]

On Thu, Mar 30, 2006 at 12:19:57PM +0200, Mike Galbraith wrote:
> Greetings,
> 
> The patch below alone makes virgin 2.6.16 usable in the busy apache
> server scenario, and should help quite a bit with other situations as
> well.
> 
> The original version helps a lot, but not enough, and the latency of
> being awakened in the expired array can be needlessly painful.  Ergo,
> speed up the array switch, and instead of unconditionally plunking all
> awakening tasks (including rt tasks, oops) into the expired array, check
> to see if the task has run since the last array switch first.  This
> leaves a theoretical hole for a stream of one-time waking tasks to
> starve the expired array indefinitely, but it deals with the real
> problem pretty nicely I think. 
> 
> For the one or two folks on the planet testing my anti-starvation
> patches, I've attached an incremental to my 2.6.16 test release.

Hmmm does not apply here :

willy@wtap:linux-2.6.16-sched25$ cat /data/src/tmp/sched_*|patch -Nsp1
willy@wtap:linux-2.6.16-sched25$ patch -p1 < /data/src/tmp/patch-2.6.16-sched-2.txt 
patching file kernel/sched.c
Hunk #1 succeeded at 71 with fuzz 2 (offset -6 lines).
Hunk #2 succeeded at 391 (offset 157 lines).
Hunk #3 FAILED at 833.
Hunk #4 FAILED at 2681.
Hunk #5 succeeded at 2770 (offset 149 lines).
Hunk #6 FAILED at 2806.
Hunk #7 succeeded at 2866 (offset 162 lines).
3 out of 7 hunks FAILED -- saving rejects to file kernel/sched.c.rej

However, it applies to plain 2.6.16 with sched_[1-6] applied (but with fuzz).
Parts of the patches are already applied by your previous patches (eg: chunk1).
I suspect that you accidentely rediffed sched.c against an intermediate
working tree instead.

For instance, this part is already provided by patch 4 :
@@ -77,6 +77,21 @@
 #define NS_TO_JIFFIES(TIME)    ((TIME) / (1000000000 / HZ))
 #define JIFFIES_TO_NS(TIME)    ((TIME) * (1000000000 / HZ))
 
+#if (BITS_PER_LONG < 64)
+#define JIFFIES_TO_NS64(TIME) \
 .../...

And this part overlaps with a previous chunk in patch 7 :

-                       rq->expired_timestamp = jiffies;
-               if (!TASK_INTERACTIVE(p) || EXPIRED_STARVING(rq)) {
+                       rq->expired_timestamp = now;
+               if (!TASK_INTERACTIVE(p) || expired_starving(rq)) {

Regards,
Willy

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/