aggregate cluster stats

Robert G. Brown rgb at phy.duke.edu
Mon Jan 7 14:40:16 EST 2002


On Mon, 7 Jan 2002, Andrew Fant wrote:

> 
> I am currently running a 118 processor cluster, with bigbrother and larrd to monitor
> system status and gather performance and utilization data on a node by node basis. 
> However, my management is now requesting aggregate statistics, and a web page
> showing load, etc, across the entire cluster.
> 
> Has anybody hacked something like this themselves?  I would rather stick close to
> bigbrother and larrd, just to simplify implementation, but I have been playing with
> SGI's open source release of PCP, and I am not adverse to switching to another
> (free) solution if it can simplify the process.

You can look at procstatd, available on http://www.phy.duke.edu/brahma.
The GUI it comes with is clunky and will soon be replaced, but the
daemon itself returns an easily parsed ascii packet, and you can chop it
up and turn it into anything you like -- realtime web page, GUI, ascii
report, graph(s).

I've been working hard on procstatd 2.0 for the last week (and hope to
finish it over the next week).  This is a near-complete rewrite that
will make it easier to parse fields, easier to add new statistics to
monitor (from /proc or elsewhere), possible to add job-specific
tracking, and possible to request a raw dump of any /proc file. (Now it
only delivers "cooked" statistics but I'm adding a command that should
allow raw proc files or lines from files to be monitored).

Accompanying the new procstatd should be a new GUI (gwatchman) in native
C and GTk that will use the new simpler and more consistent field
schema.  But be warned that although I have a functioning shell of
gwatchman cut, the guts will have to await the new procstatd API
(which it largely inspired).  It might be a few weeks before all this is
out of pre-alpha.

   rgb

> 
> Thanks for any suggestions,
>                         Andy
> 
> 

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu



_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list