In this paper, we describe the design, implementation, and modeling of a runtime kernel to support the processor farm paradigm on multicomputers. We present a general topology-independent framework for obtaining performance models to predict the performance of the start-up, steady-state, and wind-down phases of a processor farm. An algorithm is described, which for any interconnection network determines a tree-structured subnetwork that optimizes farm performance. The analysis technique is applied to the important case of k-ary tree topologies. The models are compared with the measured performance on a variety of topologies using both constant and varied task sizes. Index Terms---Parallel programming paradigms, performance evaluation, processor farm, tree networks, message passing architecture, network flow, master-slave. ------------------------------ F ------------------------------ 1I NTRODUCTION HE major problems in parallel computation revolve around questions of ease of...

The technique of the processor farm has become very widely used for parallelising applications, often being mentioned without reference to any source. The goal of this work has been to put together a complete and rigorous understanding of what the technique can be used for and what is needed in order to arrive at an efficiently farmed application. This paper consists of these two parts. We have shown, via the UNITY theory of programming, that the basic structure of the processor farm may be used to parallelise a much wider domain of applications than has generally been considered. Second, we show by example, how to build efficient implementations for the first generation of INMOS Transputers. This work is new in that it is the first that has been able to test farming harnesses by taking an abstract view of the application. This paper has been written in a semi-"instruction manual" style. Also it should serve as an introduction to the subject. 1 Introduction The processor farm was prop...