From: steven fellini (sfellini@nih.gov)
Date: 03 Dec 2007
The Helix Systems/Biowulf staff is pleased to announce the
addition of 280 nodes (976 compute cores) to the Biowulf
Cluster. Since these nodes will not immediately be part of
the general pool of nodes allocated by the batch system,
please read below for instructions on how to allocate them
for your jobs.
The new nodes are of two types:
-- Dual-socket/dual-core 2.8 GHz Opteron (4 processor),
8 GB memory, connected to gigabit ethernet (208 nodes).
Allocate with qsub property "centos".
-- Dual-socket/single-core 2.8 Ghz Opteron (2 processor),
4 GB memory, connected to Infiniband (72 nodes).
(Availability of the IB nodes will be announced separately).
Because the new nodes are running CentOS 5, an updated version
of the Linux operating system (see previous email message),
you should first test your application code with short jobs to
verify that there are no problems.
Also note that the (non-infiniband) nodes have 4 processors
per node, and except for when testing applications, you
should ensure all four processors are used by your job
('swarm' will do this automatically for you!).
To allocate the new nodes to your jobs, use the "centos"
property to qsub or swarm. For example,
% qsub -l nodes=1:centos myjob.bat
% swarm -l nodes=1:centos myswarm
Please send questions to staff@biowulf.nih.gov.