Dynamic resource management in a cluster for high-availability

Abstract

In order to execute high performance applications on a cluster, it is highly desirable to provide distributed services that globally manage physical resources distributed over the cluster nodes. However, as a distributed service may use resources located on different nodes,it becomes sensitive to changes in the cluster configuration due to node addition,reb oot or failure. In this paper,w e propose a generic service performing dynamic resource management in a cluster in order to provide distributed services with high availability. This service has been implemented in the Gobelins cluster operating system. The dynamic resource management service we propose makes node addition and reboot nearly transparent to all distributed services of Gobelins and,as a consequence, fully transparent to applications. In the event of a node failure,applications using resources located on the failed node need to be restarted from a previously saved checkpoint but the availability of the cluster operating system is guaranteed,pro vided that its distributed services implement reconfiguration features.