XSnap : a queueing network analysis package

Abstract

This dissertation describes the design and implementation of a sophisticated X-Windows based modelling package called XSnap, which can be used to solve product-form mixed multi-class queueing networks. A Graphical User Interface allows interactive network specification, whilst the modeller can also define complex network experiments and request customised output through the use of a language called SnapL. The solution modules used by XSnap are grouped together to form the Calculation Modules ToolBox (CMTB), which can be easily integrated into any modelling package which provides an appropriate user interface. Solution statistics are found using Reiser's Mean Value Analysis (MVA) algorithm, which has been extended to allow for the approximate solution of networks with PRIORITY servers or non-integral closed chain populations. A routing validation algorithm is used to validate the routing information for the network to be solved, and equations defining the relative throughput (or visit ratio) of each class at each centre in the network, are solved using a version of LU-Decomposition called Crout's method with partial pivoting. The dissertation also includes a study of a number of other available modelling packages. The choice of features included in the XSnap GUI has been largely influenced by this study. A number of different algorithms for solving product-form queueing networks are also discussed, and relevant points from this discussion are presented as part of the motivation for using the MVA algorithm for finding solution statistics.