Resource Management for Superclusters

Abstract

Clusters of workstations are now commonly found in high performance computing centers and tend to replace mainframe type supercomputers. They steadily grow in size reaching today several thousand nodes. Their complexity is also increasing: more nodes, complex network topologies, heterogeneous resources across the cluster, ...

The efficient use and administration of those big clusters, or superclusters, is not possible with currently available softwares: they lack important functionality and most of the time reliability.

This project exposes the different functionality requirements such software should meet regarding resource management: support of wide type of jobs, customizable and efficient scheduling and resource allocation, extensive monitoring, scalability, reliability, security. A priority based backfill algorithm fulfilling those requirements is proposed along with an user interface (both shell commands and graphical user interface) and some internal structures.

This page is hosted by the