[d2n-analysis-talk] Fwd: [Jlab-scicomp-briefs] JLab Data Analysis Cluster Upgrade

[Diana Parno]

Hello all,

If I'm interpreting the forwarded missive correctly, we're going to  
lose access to the 32-bit batch farm in two weeks. This could be a  
problem, since (as far as I know) our 64-bit replay is still only  
working about a fifth of the time. (Unless you've made further  
progress on that, Brad?)

If the 32-bit farm is going away, and our 64-bit replays are not yet  
working, it may be worth moving up our timetable on the five-pass  
batch of BigBite replays so that we'll have those ROOT files  
available. Of course we'd still want to push ahead on figuring out the  
64-bit replay, so that we could incorporate database adjustments in  
subsequent replay rounds, but having replayed ROOT files available  
might alleviate that time pressure.

Thoughts?

Best,
Diana

Begin forwarded message:

> From: Sandy Philpott <sandy.philpott at jlab.org>
> Date: February 28, 2011 12:45:59 PM PST
> To: jlab-scicomp-briefs at jlab.org
> Cc: "scicomp at jlab.org" <scicomp at jlab.org>
> Subject: [Jlab-scicomp-briefs] JLab Data Analysis Cluster Upgrade
> Reply-To: Sandy.Philpott at jlab.org
>
> Jefferson Lab's experimental physics data analysis cluster has  
> recently been upgraded with the FY11 Physics farm node procurement.   
> The new systems are all configured with 64-bit CentOS 5.3.
>
> The interactive systems will gain an AMD Opteron 32-core node named  
> ifarm11001 ("interactive farm, FY11 system 001") on Monday, March  
> 7.   The new system is currently available as ifarmltest  
> ("interactive farm, linux test"), with 64GB RAM and a 1TB disk  
> including local /scratch.  This node is being added to the current  
> ifarml64 pool, consisting of ifarml1 and ifarml6,  two 8-core  
> hyperthreaded Nehalem systems.
>
> The batch farm is already running 22 additional nodes,  
> farm11001-022. These include 18 Nehalem 8-core hyperthreaded systems  
> with 24GB RAM and 1TB disk, and 4 AMD Opteron 32-core systems with  
> 64GB RAM and 1TB disk, for a total of 272 additional cores.  The 32- 
> core AMD systems are being configured to run multithreaded jobs,  
> rather than 32+ single core jobs.  Multithreaded jobs are identified  
> by the cpu: n tag in the job submission.  This addition brings the  
> total number of 64-bit cores in the farm to just over 700.
>
> With this addition the 32-bit end-of-life Fedora nodes are in the  
> process of decommissioning as planned. These include the interactive  
> nodes ifarml3, ifarml4, ifarml5, and the last batch of 25 dual core  
> farm systems.  ifarml2, an even older system, will also be  
> decommissioned. These old nodes will all be shut down on Monday  
> March 14.
>
>
> _______________________________________________
> Jlab-scicomp-briefs mailing list
> Jlab-scicomp-briefs at jlab.org
> https://mailman.jlab.org/mailman/listinfo/jlab-scicomp-briefs