[Jlab-scicomp-briefs] New JLab farm14 compute nodes: CentOS 6.5, ~2500 cores

[Sandy Philpott]

Hello and Happy Holidays! 


The new farm14 compute nodes are in service and have been heaviily tested by a few users. Early in 2015 we will move these Haswell CentOS 6.5 nodes into production service, and provide ifarm interactive farm user login nodes for batch job preparation. Then later the current production farm nodes will be upgraded from CentOS 6.2 to 6.5. 


See these links for some system views of these new nodes: 
https://scicomp.jlab.org/scicomp/#/ choose Node Status 
https://scicomp.jlab.org/scicomp/#/operations/nodes 
https://scicomp.jlab.org/ganglia/ choose PBS Farm Cluster 



Please let us know via email to farm at jlab.org if you'd like an account for testing/running on these new systems before the holidays. All feedback welcome to farm at jlab.org and your Physics Hall Computing Coordinators ( https://scicomp.jlab.org/docs/node/31). 


Regards, 
Sandy 

----- Original Message -----

From: "Sandy Philpott" <philpott at jlab.org> 
To: jlab-scicomp-briefs at jlab.org 
Sent: Thursday, November 20, 2014 12:00:55 PM 
Subject: Testing new farm14 CentOS 6.5 nodes 


Hello Farm Users, 


The first batch of farm14 nodes are almost ready for early testing; we're working out a few final details. The systems are built at CentOS 6.5, as the new hardware required a newer kernel than our existing farm CentOS 6.2 systems. 


If you're ready now to try running your batch jobs in the updated environment, please send an email to farm at jlab.org, and we'll work with you to get started. 


Sandy 

----- Original Message -----

From: "Sandy Philpott" <philpott at jlab.org> 
To: jlab-scicomp-briefs at jlab.org 
Sent: Tuesday, November 11, 2014 1:28:14 PM 
Subject: New farm14 nodes adding ~2500 cores 







JLab Scientific Computing Users, 

The 104 new Physics nodes for the data analysis farm cluster have just arrived (well, except their power cords, coming tomorrow). These farm14* nodes are dual Intel E5-2670v3 (Haswell) 12 core, 2.3 GHz, with 32 GB DDR4-2133 memory -- the leanest memory configuration in the farm (less than 1 GB per job slot on serial jobs). They will run with as many slots as experience shows they can handle, based upon statistical measurements in actual running. Their network includes 20 gigabit Infiniband DDR for I/O to the Lustre and ZFS filesystems. Their initial installation in the F112 Data Center begins this week. 


This farm node installation will add a total of 104 * 24 = 2496 cores to the existing 1400 core farm. We will be working with early users to identify the correct operating system and kernel to install, then move the systems to production mode as their burn-in proves successful. More details to follow. 


Regards, 
Sandy 





-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.jlab.org/pipermail/jlab-scicomp-briefs/attachments/20141219/25ab8af3/attachment-0003.html>