Adaptive load balancing and multithreading

In this paper we address adaptive load balancing, multithreading and granularity. Firstlty, we describe an Adaptive Load Balancing Algorithm called ALBA. This latter includes a load information collection policy that uses adaptive collection frequencies which initial and lower values are computed according to the underlying architecture and programming environment. Moreover, ALBA integrates a mean which prevents from flooding of nodes. Secondly, we present the task-based and thread-based approaches. The second approach has been retained for the implementation of ALBA on a farm of DEC/ALPHA processors. Thirdly an application to a muItithreaded IDA* program applied to the 15-puzzle problem is presented with some interesting preliminary results. We also propose an approach for managing the granularity issue within the framework of the above application.