On Feb 22, 2011, at 1:22 AM, Jerome Flesch wrote:
>> A scheduler quantum of 10ms (or HZ=100) is a common granularity; probably 
>> some other process got the CPU and your timer process didn't run until the 
>> next or some later scheduler tick.  If you are maxing out the available CPU 
>> by running many "openssl speed" tasks, then this behavior is more-or-less 
>> expected.
>> 
> 
> We did most of our tests with kern.hz=1000 (the default FreeBSD value as far 
> as I know) and we also tried with kern.hz=2000 and kern.hz=10000. It didn't 
> change a thing.

For a long time kern.hz=100 was the default; more recent versions have switched 
to kern.hz=1000, which is beneficial for minimizing latency for things like 
ipfw/dummynet processing, but also involve greater scheduler overhead.  
kern.hz=10000 is likely to reduce performance and may have odd effects upon 
certain kernel timers.

> Also, we are talking about a process not being scheduled for more than 100ms 
> with only 1 instance of openssl on the same CPU core. Even with a scheduler 
> quantum of 10ms, I find that worrying :/

It depends on what else the machine is doing.  Gathering some data via acct/sa 
might be informative, as you might be running some other tasks via cron or 
whatever which your testing isn't expecting.

> We expected both processes (the test program and openssl) to have each half 
> the CPU time and being scheduled quite often (at least once each 10ms). 
> According to the output of our test program, it works fine for most of the 
> calls to clock_gettime(), but from time to time (about 1 loop in 200000 on my 
> computer), we have a latency pike (>= 100ms).
> 
> Thing is, these pikes wouldn't worry us much if they wouldn't last longer 
> than 1s, but they do on some occasions.

Also, are you sure that you don't have ntpdate or ntpd or something else 
adjusting your system clock underneath you...?

Regards,
-- 
-Chuck

_______________________________________________
freebsd-current@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscr...@freebsd.org"

Reply via email to