Ares achieved a whopping 250 Gflops peak rate performance as measured by the HPL parallel linear algebra benchmarking code. A 64 thousand by 64 thousand dense system of equations was solved in only 782 seconds! Ares would have qualified within the first 100 fastest supercomputers in the world, … in the year 2000, according to www.top500.org,

Filed under: Cluster Links | Comments Off

The easiest way to set up a parallel job is to use mpi directly. For that a file named, say, machines should list the nodes on which you wish to run your job; each line contains the name of the node and the number of slots (cores) available for alocation. For example:

compute-0-0 slots=8
compute-0-1 slots=8

If you use a scheduler such as sge, you don’t need to care about the allocation of the machines, since that would be scheduler’s job.

Then: 

(1) compile your program: mpicc -o my_prog my_program.c
(2) set the shell for ssh not to use passwords: ssh-agent $SHELL
(3) authenticate once for all: ssh-add
(4) run your program on n slots: mpirun -np n -hostfile machines my_prog

Filed under: recipies | Comments Off

This cluster is managed by Rocks, an open-source Linux cluster distribution that makes building and using clusters for scientific computation very easy. The National Science Foundation is the main funding vehicle for the Rocks project.

Click on the link on the right to see live details about the cluster.

Filed under: Cluster Links | No Comments »