On 8/3/07, Ditesh Kumar <[EMAIL PROTECTED]> wrote:
> On Thu, 2007-08-02 at 18:56 +0800, Ang Chin Han wrote:
> > web{redacted}:~ # uptime
> > 4:51pm up 10 days 4:43, 4 users, load average: 67.60, 68.22, 61.98
>
> this is on the same xeon 1 core boxen?
Nope. Another production box for a client.
> did some reading on what uptime really means. apparently, depending on
> the system its on, it can mean different things. on linux boxen,
Nitpicking. Uptime is very specific, itym "load avg". The command
"uptime" just happens to display the load average.
So long as I remember, load average means number of processes in
runtime *or* uninterruptible sleep, for the duration of 1, 5 and 15
mins.
Hmm, http://en.wikipedia.org/wiki/Load_(computing) describes pretty
much it, not really that ambiguous.
> sleeping processed blocked on i/o count towards load average, which may
> or may not be accurate depending on what you think load average should
> mean. so apparently, a responsive boxen with load avg of 70 is quite
> feasible in such scenarios :)
>
> my boxen which was not responsive on load avg of 17 because it was cpu
> bound as opposed to being i/o bound.
In wonder if renicing postmaster or apache for that matter to 10 would
be a good idea to allow at least a good shot at allowing admins to
ssh-ing in to fix things.
off-topic pgsql tip: turn on stats_start_collector,
stats_command_string, stats_block_level and stats_block_level in
postgresql.conf. You'll take a leeeetle performance hit, but "select *
from pg_stat_user_tables "and pg_stat_user_indexes and
pg_stat_activity will give you some hints on where pgsql spends most
of its time.
_______________________________________________
Slugnet mailing list
[email protected]
http://www.lugs.org.sg/mailman/listinfo/slugnet
