Hi, I suspect a CPU bottleneck in one of our PostgreSQL servers but not sure how to confirm the suspect. It's a DELL Box running CentOS 5.4 with 64GB RAM and 16 XEON E7430 2.13 GHz processors. vmstat r column "run queue" usually indicates values higher than 2 and less than 5 but "Load Average" values from top, sar -q and other commands show always values less than 1. Should not these values be higher than 16 on a box with 16 processors to confirm a CPU constraint? Thank you in advance! -- Reimer 47-3347-1724 47-9183-0547 msn: carlos.reimer at opendb.com.br
On Sat, Feb 09, 2013 at 04:24:19PM -0200, Carlos Henrique Reimer wrote:> processors. vmstat r column "run queue" usually indicates values higher > than 2 and less than 5 but "Load Average" values from top, sar -q and other > commands show always values less than 1. > > Should not these values be higher than 16 on a box with 16 processors to > confirm a CPU constraint?If you have single-threaded processes then that process could chew up 100% of a single core but not be able to run any faster. The load average would only be 1. Load average is a poor measure; look at %idle on each core. If you see one core with 0% idle then something is maxed out. -- rgds Stephen
Hi, All 16 core show always %idle greater than 95% (mpstat -P ALL). So can I assume there is no CPU constraint in this box? Thank you! On Sat, Feb 9, 2013 at 4:37 PM, Stephen Harris <lists at spuddy.org> wrote:> On Sat, Feb 09, 2013 at 04:24:19PM -0200, Carlos Henrique Reimer wrote: > > processors. vmstat r column "run queue" usually indicates values higher > > than 2 and less than 5 but "Load Average" values from top, sar -q and > other > > commands show always values less than 1. > > > > Should not these values be higher than 16 on a box with 16 processors to > > confirm a CPU constraint? > > If you have single-threaded processes then that process could chew up 100% > of a single core but not be able to run any faster. The load average > would only be 1. > > Load average is a poor measure; look at %idle on each core. If you see > one core with 0% idle then something is maxed out. > > -- > > rgds > Stephen > _______________________________________________ > CentOS mailing list > CentOS at centos.org > http://lists.centos.org/mailman/listinfo/centos >-- Reimer 47-3347-1724 47-9183-0547 msn: carlos.reimer at opendb.com.br