[olug] Linux Journal - Nov 2010 - LHC

Carl Lundstedt clundst at unlserve.unl.edu
Thu Oct 21 02:06:35 UTC 2010

  On 10/20/2010 5:26 PM, Kevin D. Snodgrass wrote:
> --- On Wed, 10/20/10, Carl Lundstedt<clundst at unlserve.unl.edu>  wrote:
>> Linux Journal has made this article
>> free to read:
>> http://www.linuxjournal.com/content/the-large-hadron-collider
>> Cheers,
>> Carl Lundstedt
>> UNL
> I love it!  "PhEDEx"
> One thing, in the article you mention some software analogous to "SETI at home" (or folding at home, which I used to run) called OSG, but for university computing centers.  Has anyone cosidered extending that to an actual public client, i.e. LHC at home or maybe HiggsBoson at home?  (GodParticle at home?)
> Kevin D. Snodgrass
OSG is used to share the resources of a campus cluster.  It uses globus 
and a pretty heady middleware stack.  It presumes a Linux environment, a 
place to install software and  that the job will basically own and not 
be evicted from.  OSG is for dedicated sites.

There's something out there called Einstein at home for scientists to get 
cpu hours off volunteer machines.  We sometimes get jobs that start 
einstein at home on our workers, do their work, then leave.  Einstein at home 
is getting some use.

There is also an LHC at home (http://lhcathome.cern.ch/), but the CMS (my 
experiment) software stack is really complex (both because of what it is 
and because of who wrote it,  physicists aren't the best coders).  Right 
now a user analysis job assumes that it will have access to 1.5 GB of 
ram, 10 GB of disk, that's a little much to ask of volunteers.  That's 
not the real challenge though.  A standard CMS data file is 2GB, so 
there's a ton of bandwidth required in moving the data around.  That's a 
tall order to ask from both the volunteer and the experiment.  I know 
I'd hate to serve a file to a machine and have the job killed before it 
could finish.  The opportunity for waste is currently too high (IMO).

Right now the computing model has the jobs chasing the data.  There are 
efforts to invert the model and have the data chase the cpu resources.  
IF that inversion takes place it might be time to look at trying to use 
commodity, volunteer cpus.

Thanks for the interest,

More information about the OLUG mailing list