User-profile adaptable resource management system for workstation cluster architectures

摘要

The increasing use of workstation clusters as a substitute for distributed memory computing systems has gained widespread popularity due to the existence of the inexpensive and powerful workstations. Besides using remote execution commands in a workstation cluster users have incomplete information about the state of the whole system. Therefore, a large number of workstations frequently remain idle or lightly loaded where others are heavily loaded. Existing job management systems assign jobs to idle or most lightly loaded workstations according to the state of the underlying resources. Therefore the impact by interactive users on remote workstations is relatively high. We introduce a new concept of a resource management system which allocates the best suited workstation(s) to the user according to his/her former activities (User-Profile). By using this concept, there is no impact by interactive users because workstations have no owners but the workstation pool is public among all users. The possibilities of applying this concept on a traditional workstation cluster architecture and a new workstation cluster with a concurrent network architecture (CNA) is introduced too.