Networks of workstations (NOWs) are becoming increasingly popular for providing cost-effective and affordable parallel computing. However, the LAN environments are gradually becoming heterogeneous due to the commodity nature of workstations and networking equipment. Such heterogeneity gets reflected in terms of communication capabilities of nodes (processor speed, memory speed, network interface support) and network architectures (different interconnects with varying topology and connectivity). This is forcing NOW environments to be gradually redefined as `Heterogeneous Networks of Workstations' (HNOWs). This research is targeted towards providing fast and scaleable collective communication support for such emerging HNOWs. The following problems are being addressed by this project: 1) quantitatively specifying the heterogeneous characteristics of a HNOW environment and studying their impact on the performance of current implementations of collective communication operations; 2) designing and developing optimal algorithms for collective communication operations by considering major heterogeneous characteristics; and 3) studying the performance gains achieved by the optimal algorithms for different suites and HNOW environments. An experimental HNOW testbed consisting of a set of workstations with different speeds and different interconnect types (Fast Ethernet, ATM, and Myrinet) is planned to be used together with theoretical, analytical, and simulation approaches to develop and validate the new research results.
