PERFORMANCE BOUNDS FOR DISTRIBUTED SYSTEMS WITH WORKLOAD VARIABILITIES AND UNCERTAINTIES

Bounding techniques for queuing network models used to analyze the per formance of parallel and distributed computer systems accept single va lues as model inputs. Uncertainties or variabilities in service demand s may exist in many types of systems. Using models with a single aggre gate mean value for each parameter for such systems can lead to inaccu rate or even incorrect results. This paper proposes to use histograms for characterizing model parameters that are associated with uncertain ty and/or variability. The adaptation of the well-known asymptotic bou nds as well as balanced job bounds for single class queuing networks t o histogram parameters is presented in the paper.