The Implementation of a Scalable Gang Scheduling Scheme on the AP1000+

Some disadvantages associated with conventional gang scheduling are that it is not scalable and that space-sharing policies are not easily incorporated to enhance the efficiency of resource utilisation due to the centralised control and the frequent signal-broadcasting across the processors. This paper describes an effective and scalable gang scheduling scheme and its implementation on the Fujitsu AP1000+. 1 Introduction With the rapid developments in both hardware and software technology the performance of scalable systems such as networks of workstations, Boewolf system (clusters of PCs) and clusters of SMPs have significantly been improved. It is expected that this kind of system will dominate the parallel computer market in the near future because of the continued cost-effective growth in performance. There is no doubt that some of these machines will be dedicated to specific applications such as grand challenges. However, a great number will be used as generalpurpose computing se..