Impact of workload model on PACS performance prediction

摘要

Abstract: Discrete event simulation is often used to predict the performance of PACS systems. Although detailed models can be developed for the system elements, the image generation workload assumptions can have a significant impact on the results. The selection of a source model is a crucial part of any simulation and should be representative of the real application. If the source model is improperly selected it is possible to reach misleading conclusions. Examples of source models that have been used for the image workload in PACS include Poisson, modulated Poisson, and bursty. In this paper we compare the use of different image generation models and the impact on system simulation results. The BONeS simulation package is used to define a PACS system consisting of medical imaging device sources and destinations (archives and displays) which exchange images over a network consisting of ethernet and token bus subnetworks. All parameters ar identically specified for each simulation experiment with the only difference being the type of traffic generator used. The results illustrate that use of a Poisson model results in an overly optimistic estimate of system behavior. An alternative model is proposed which defines imaging device traffic patterns in terms of bursts of images within `studies.' The parameter sensitivity of this model is also examined. !14