On Thu, Sep 10, 2009 at 1:40 AM, Raymond Wong <[email protected]> wrote:
>
> Hi,
>
> I am experiencing poor performance on our Solaris 10 installation.
>
> The bash script that I wrote took more than 20 minutes to complete on the 
> Solaris instance, while the exact same script only take 45 seconds on an 
> intel-linux installation.
>
> Though the Intel CPU is running at 2.8GH & the Sparc at only 1GH. That 
> doesn't account for the huge time difference. Both systems are multiple CPUed 
> & with normal load.
>
> Any ideas what might be causing the poor performance?

Others have mentioned that there are a lot of process creation that
can be avoided.  Part of process creation is managing memory mappings.
 On a multi-processor system, updating memory mappings requires cross
calls to other processors to be sure that they don't have stale data
in cache and/or the memory management unit.  The specific chip and
chipset architecture (multi-socket, multi-core, multi-thread, on-cpu
MMU, etc.) will likely impact the relative impact of cross calls.

Much like there is an upper limit on the number of iops for a disk, I
believe (but have not rigorously measured) that there is an upper
limit on the number of process creations per second.  I expect that
the number of process creations per second will degrade (potentially
in a far worse than linear form) with the number of processors.  Does
anyone have some hard data to support or refute this?

--
Mike Gerdts
http://mgerdts.blogspot.com/
_______________________________________________
sysadmin-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/sysadmin-discuss

Reply via email to