No abstract

This pqxr introduces a family of interconnection networks for loosely-coupled multiprocessors called Hierarchical Cubic Nmorks (HCNs). HCNs use the well-known hypercube network as their basic building block. Using a considerably lower number of links per no&. HCNs realize lower network a7~er.r than the hypercube. The pe@rmance of several well-known applications on a hypothetical system employing the HCN is identical to their performance on a hypercube. HCNs thus enjoy the same advantages as a hypercube, albeit with considerably siinpler interconnections.

No abstract

The relatively maturefiber optic technology, enhanced by the use of wavelength division multiplexing (WDM), provides a solution for surmounting the problems associated with copper based interconnections for multiprocessors as the CPU clockfrequencies increase. In this paper we undertake a trace-drivensimulationstudy of theperformance of several fast channel access schemes for WDM channels that are appropriate for multi-dimensional interconnections for multiprocessors with thousands of nodes. We show that the protocols that employ tunable components cannot provide adequate performance unless the tuning times improve to 500 nanoseconds. The main conclusion of our study is that a simple WDM channel access protocol, called CLUSTER, based on the use offied-tuned optical transmitters and receivers, is the protocol of choice for building multidimensional WDM interconnections. The recent advent of single chip laser and receiver arrays for WDM links, employing fixed-tuned components, provides an added incentive for using the CLUSTER protocol for multidimensional WDM interconnections for multiprocessors.

No abstract