On 11/4/07, jpd <[EMAIL PROTECTED]> wrote:
>    PID USERNAME  SIZE   RSS STATE  PRI NICE      TIME  CPU
> PROCESS/NLWP
>   5496 root      298M   78M run     49    0   0:02:20 3.8% hlxserverplus/11
>   5900 webservd  163M   45M sleep   59    0   0:00:09 1.9% httpd/1
>   5884 webservd  166M   43M sleep   59    0   0:00:06 0.0% httpd/1
>   5887 webservd  163M   41M sleep   59    0   0:00:10 0.0% httpd/1
>   5906 webservd  159M   35M sleep   59    0   0:00:01 0.0% httpd/1
>   5888 webservd  165M   34M sleep   59    0   0:00:04 0.0% httpd/1
>   5885 webservd  160M   33M sleep   59    0   0:00:03 0.0% httpd/1
>   5886 webservd  160M   29M sleep   59    0   0:00:03 0.1% httpd/1
>   5902 webservd  160M   29M sleep   59    0   0:00:05 0.0% httpd/1
>   5905 webservd  160M   21M sleep   59    0   0:00:01 0.0% httpd/1
>   5878 root      158M   16M sleep   59    0   0:00:01 0.0% httpd/1
>  ......
> ZONEID    NPROC  SWAP   RSS MEMORY      TIME  CPU
> ZONE
>      5       37  270M  121M   9.5%   0:01:12 2.1%
> www
>      4       35  330M   80M   6.3%   0:03:58 3.8%
> helix
>      0       49   68M   93M   7.3%   0:37:10 0.8%
> global
>      3       28   37M   53M   4.1%   0:23:21 0.7% icecast1
>
> so according to prstat apache is still using about 1g of swap+memory and
> more RSS that the zones is?

But you forget that many processes will have shared mappings.  If a 10
MB file is mapped by 100 processes, that is only 10 MB used, not 1 GB.
 The sum of the swap reservation size from pmap -S <all pids> would
likely be more in line with the value used for the swap resource
control.

In the not too distant past, prstat would naively sum the RSS of all
the processes in the zone and say that was the RSS for the zone.  This
can gave ridiculous results when you had something like oracle with an
hundreds of processes mapping an SGA (shared memory segment) that was
many gigabytes in size.  I've seen large systems say that tens of
terabytes of RAM were in use when the systems (obviously) had less
than a terabyte of RAM.

You can also use "vmstat" and "vmstat -p" to confirm that paging
activity is high and somewhat characterize what is being paged.

-- 
Mike Gerdts
http://mgerdts.blogspot.com/
_______________________________________________
zones-discuss mailing list
zones-discuss@opensolaris.org

Reply via email to