Information Collection Policies: Towards load balancing of communication-intensive parallel applications

In communication-intensive applications, the information collection policy influences the load balancing decision process and the communication itself, because the collection reduces the bandwidth used by the application. In this paper, we present a study of collection policies used by well-known load balancing system, now in the context of communication-intensive applications and focusing in response times (time of reaction against instabilities) and bandwidth. We divided the policies in server and broadcast oriented, which can share all of partial load information, implementing them with an asynchronous communicated middleware. Our experimental results show a better performance (in response time and bandwidth) of broadcast oriented policies sharing partial information.