Re: PROBLEM: high load average when idle

[Linus Torvalds <torvalds@linux-foundation.org>]

On Wed, 3 Oct 2007, Chuck Ebbert wrote:
> 
> But we reduce the number of samples because some ticks just never
> happen when the timers get rounded:
> 
> No rounding:
> 
>       tick ............... tick
>      1 running            1 running
> 
> Rounded:
> 
>       tick
>      2 running
> 
> In the first case the average is 1, but it's 2 in the second.

In fact, I think this is it!

The load average is not calculated every tick, because that's not just 
expensive, but we also want to have some time-based decay. So it's 
calculated every LOAD_FREQ ticks.

And guess what: LOAD_FREQ is defined to be exactly five seconds.

So imagine if the timer gets to be in sync with another event that happens 
every five seconds - let's pick at random a 5-second JBD transaction 
thing?

Anders - does this idiotic patch make a difference for you?

Without this, I can easily imagine that the rounding code tends to try to 
round to an even second, and the load-average code generally also runs at 
even seconds!

		Linus

---
 include/linux/sched.h |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/include/linux/sched.h b/include/linux/sched.h
index a01ac6d..643de0f 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -113,7 +113,7 @@ extern unsigned long avenrun[];		/* Load averages */
 
 #define FSHIFT		11		/* nr of bits of precision */
 #define FIXED_1		(1<<FSHIFT)	/* 1.0 as fixed-point */
-#define LOAD_FREQ	(5*HZ)		/* 5 sec intervals */
+#define LOAD_FREQ	(5*HZ+1)	/* ~5 sec intervals */
 #define EXP_1		1884		/* 1/exp(5sec/1min) as fixed-point */
 #define EXP_5		2014		/* 1/exp(5sec/5min) */
 #define EXP_15		2037		/* 1/exp(5sec/15min) */