ICT has recently added a 512-core HPC Linux compute cluster, Plato.usask.ca, to the High Performance Computing resources on campus. Plato is specifically targeted for research use. It is designed to complement the HPC resources available in Compute Canada/WestGrid, and those provided by the ICT cloud service.
Plato enables U of S researchers to run research problems that require an extended period of compute time, very large memory allocations, and/or many processors at once. The cluster is designed to enable the research of a wide variety of groups: for example, newly arrived researchers who want to hit the ground running; experienced users who need a short turnaround time for results; and users whose need for HPC is less frequent, but still important. Plato is already increasing research productivity by reducing time waiting for results, and enabling researchers to tackle problems of a larger scope than would have been otherwise feasible, while making HPC available to more researchers and graduate students.
There currently is no charge for the use of Plato for U of S research purposes.
To ensure that all users get equitable access to the resources of Plato, a system is in place to do fair sharing. The “fairshare” system uses a batch system similar to the one used on WestGrid and Compute Canada clusters. Users are grouped by the responsible faculty member so that any resources are distributed equitably according to their “fairshare.”
More detail on the fairshare implementation on Plato, as well as all the technical information, can be found on the Research Computing website.
A number of researchers have been putting Plato through its paces to ensure that installed software (and hardware) is working optimally. So far, the performance has been reported by researchers to be excellent (one user writes “I must say that Plato is pretty fast!”).
The scheduling and operation policies for Plato will be reviewed by the High Performance Computing Advisory Committee in Fall 2013.
For access to Plato (or questions about it or any of the other HPC systems, including Socrates, Moneta, and Zeno), please contact the High Performance Computing team at firstname.lastname@example.org.