Distributed System Model

 

Earlier simulations such as those used by [ELZ86b, LO86, Zho87, BMD94, HBD95], tended to accurately represent only the CPU resource in the system, and while adequate for those studies, evaluation of the process mix hypothesis requires a full set of resources to be modelled.

As load distribution is concerned with the interaction of a set of processes and compute servers, these two components obviously form the basis of a model for the distributed system simulation.

• A Set of Processes: Each process is characterised by the set of resources it requires to finish and the rate at which each resource is consumed.
• A Set of Compute Servers: Each compute server consists of a finite set of resources for consumption by processes.

There are two possible approaches to constructing a model from these components. Firstly the processes could be passive, and the compute servers the active component, dividing its resources evenly between its processes, paying no attention to the rates at which different process consume different resources. A more realistic approach is for the the processes to be active, driving the simulation with their processing demands, leaving the compute servers as simple providers of system resources and arbitrators of contention.

This leads into an explanation of the structure of the chapter. As processes are the active agents in the simulation, the first half of the chapter explains the modelling of the processes, and the way in which they flow between resources. The second half of the chapter details the simulation of the compute server, and its service rates.