|
Abstract:
Bounding techniques used for evaluating the performance of parallel and distributed computer systems accept single values as model inputs. Existence of uncertainties or variabilities in service demands may exist in many types of systems making the use of a single mean value for each model parameter inappropriate and causing the conventional modelling approach to become ineffective. Thus, the conventional bounding techniques are not directly applicable to systems with variabilities and uncertainties in workload. Although the clients in a client-server system are statistically identical, factors such as the time of the day and the current size of the data files can introduce variabilities in service demands for the server devices. Such a system cannot be analyzed by a multiclass bounding technique that is concerned with dissimilar client classes. This paper proposes to use histograms for characterizing one or more model parameters that are associated with uncertainty and/or variability. The adaptation of the existing asymptotic bounds as well as balanced job bounds to histogram parameters is presented in the paper. One or more input parameters for the model can be specified as a histogram. It is shown that not considering observed variabilities may lead to highly inaccurate and incorrect bounds on system performance. Examples applying these techniques to different problem domains are discussed and an illustrative example modeling the client-server architecture of the computer system of a telemarketing company is presented in detail.
|
