Work related to the implementation of process farms has been done on transputer based environments which are typically distributed memory MIMD message passing environments. The work done at EPCC [2] on process farming runs on top of their CHIMP environment which also works as a heterogenous computing platform using workstation clusters. Their implementation of farming however consists of a source process and a sink process (instead of a single farmer task). Implementations of static farms with queue lengths (of the local worker task's queue of work packets) specified by worker tasks have been done at C-DAC, India [4] and TUL, India [3] but have concentrated on a homogeneous environment. The advantage of [3] over this paper's implementation is that the farm related information is maintained in a distributed manner, so when multiple farms are implemented it would have no bottlenecks at the central server site. This may happen in our case if all farms simultaneously dispatch work packets to the workers. Work on a single process farm (non hierarchical) based application implementation also appears in 3L[6] in Inmos, U.K.
The author is not aware of any process farm implementation under the PVM heterogenous computing environment to implement dynamic load balancing. This work aims at bridging this gap by providing a dynamic load balancing construct to expand the PVM API.