[Lustre-discuss] LMT 2.6.0

[Jim Garlick]

If the goal is to get lustre data into ganglia and use existing 
mechanisms to look at the data, LMT is probably not the right path.

But if you wanted to set up LMT and just replace the back-end data
collection mechanism, that is probably doable.  I'm afraid my 
knowledge of ganglia falls short here but I would assume you could
craft a shell script that feeds lustre /proc values to gmetric
or write some sort of "native" ganglia module to do the job more 
efficiently.

Jim

On Wed, Apr 09, 2008 at 11:57:01AM -0700, Klaus Steden wrote:
> 
> Hi Jim,
> 
> Hmmm, good to know ... does LMT support shell call-outs? If so, it would be
> easy enough to have it populate a Ganglia receptor using gmetric (using the
> older pre-3.1.x framework) ... not sure about recording from Ganglia to
> mysql, although it would record historical data using rrdtool.
> 
> Klaus
> 
> On 4/9/08 11:32 AM, "Jim Garlick" <garlick at llnl.gov>did etch on stone
> tablets:
> 
> > Hi Klaus,
> > 
> > Nobody's done that yet that I know of, but it would seem doable.
> > 
> > Cerebro is similar to ganglia in that it uses multicast with
> > listeners and talkers.  You would need machinery to get lustre proc
> > values into ganglia on the lustre servers (the talkers), and then from
> > ganglia to the mysql database (the listener).
> > 
> > Jim
> >