EIA-9806741 Grama, Ananth Y. Park, Kihong Purdue University ESS: ISAC: Integrated System Support for Adaptive Communication and Computation Control in Clustered Environments This project aims to build an integrated software system (ISAC) for application development in cluster-computing environments. ISAC addresses issues of communication control and load balancing to facilitate an effective parallel platform. In contrast to traditional approaches, ISAC views communication and computation control in an integrated system. It dynamically schedules computation and communication making use of their repetitive nature to achieve optimal performance. End-to-end congestion control is augmented by a group communication mechanism that conjointly schedules communication and computation. Scheduling is carried out assuming a best-effort network environment in which resources are shared among a pool of applications. Communication control is adaptive, and scheduling is performed dynamically as a function of network state. Assuming that routers implement a form of weighted fair queuing wherein resource reservations can be affected, ISAC implements guaranteed-bandwidth channels either per-connection or per-multicast group using a form of admission control. In conjunction with process scheduling support at host stations, this allows critical applications to achieve guaranteed progress impervious to network and processor loads. It also allows applications to deal with heterogeneity among hosts and networking environments transparently. ISAC will significantly enhance the programmability and performance of cluster-computing applications ranging from scientific computing to distributed web and database servers.
