load average wraps at 1024

load average wraps at 1024

[Jesper Juhl]

Hi,

Don't know if this is a kernel issue or userspace issue, but load
average values wrap back to 0 once they hit 1024.

I've been trying to stress 2.6.15-mm3 by putting a *lot* of load on it
and seeing if it stays alive, how it copes, how long it seems to take
to recover etc.
While doing that I've done some test runs that start thousands of
processes and the load average quickly shoots up to several hundred
and eventually reach 1000 - when it continues to climb it goes to 1023
and then wraps down to small numbers like 4-5 and then continue
climbing from there. Once I kill all my processes it slowly goes down
to zero, then wraps back to ~1000 and continues to climb down from
there until it's eventually back to normal.

Is this expected behaviour?

--
Jesper Juhl <jesper.juhl@gmail.com>
Don't top-post  http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please      http://www.expita.com/nomime.html

Re: load average wraps at 1024

[Nish Aravamudan]

On 1/11/06, Jesper Juhl <jesper.juhl@gmail.com> wrote:
> Hi,
>
> Don't know if this is a kernel issue or userspace issue, but load
> average values wrap back to 0 once they hit 1024.
>
> I've been trying to stress 2.6.15-mm3 by putting a *lot* of load on it
> and seeing if it stays alive, how it copes, how long it seems to take
> to recover etc.
> While doing that I've done some test runs that start thousands of
> processes and the load average quickly shoots up to several hundred
> and eventually reach 1000 - when it continues to climb it goes to 1023
> and then wraps down to small numbers like 4-5 and then continue
> climbing from there. Once I kill all my processes it slowly goes down
> to zero, then wraps back to ~1000 and continues to climb down from
> there until it's eventually back to normal.
>
> Is this expected behaviour?

I'm going to say yes, based on the comments in sched.h:

/*
 * These are the constant used to fake the fixed-point load-average
 * counting. Some notes:
 *  - 11 bit fractions expand to 22 bits by the multiplies: this gives
 *    a load-average precision of 10 bits integer + 11 bits fractional
 *  - if you want to count load-averages more often, you need more
 *    precision, or rounding will get you. With 2-second counting freq,
 *    the EXP_n values would be 1981, 2034 and 2043 if still using only
 *    11 bit fractions.
 */

That is, 10 bits of integer is 1024, the remaining 11 bits being used
for fractions.

Thanks,
Nish