Charles H. Sauer

Approximate analysis of central server models

Service time distributions at computer processing units are often nonexponential. Empirical studies show that different programs may have markedly different processing time requirements. When queuing disciplines are first come, first served, preemptive priority or nonpreemptive priority, models reflecting these characteristics are difficult to analyze exactly. Available approximate techniques are often too expensive for parametric analysis. Inexpensive approximate techniques for solution of central server models with the above characteristics are presented. Ther esults of these techniques are validated with simulation results.

Approximate solution of queueing networks with simultaneous resource possession

Queueing networks are important as Performance models of computer and communication systems because the performance of these systems is usually principally affected by contention for resources. Exact numerical solution of a queueing network is usually only feasible if the network has a product form solution in the sense of Jackson. An important network characteristic which apparently precludes a product form solution is simultaneous resource possession, e.g., a job holds memory and processor simultaneously. This paper extends previous methods for approximate numerical solution of queueing networks with homogeneous jobs and simultaneous resource possession to networks with heterogeneous jobs and simultaneous resource possession.

Computational algorithms for state-dependent queueing networks

Queueing networks are important as performance models of computer and communication systems. Exact numerical solution of a queueing network is usually only feasible if the network has a product form solution in the sense of Jackson. Product form networks allow a rich variety of forms of statedependent behavior. However, efficient computational algorithms have not been developed for several of the allowed forms of state-dependent behavior. This paper develops the two most important computational algorithms. Convolution and Mean Value Analysis, to apply to forms of state-dependent behavior allowed in product form networks. It is demonstrated that these two algorithms are of equal generality, contrary to common belief.

The research queueing package: past, present, and future

Queueing networks are important as performance models of systems where performance is principally affected by contention for resources. Such systems include computer systems, communication networks, office systems and manufacturing lines. In order to effectively use queueing networks as performance models, appropriate software is necessary for definition of the networks to be solved, for solution of the networks (by numerical, approximate and/or simulation methods) and for examination of the performance measures obtained. One of the most widely known and influential pieces of queueing network software is the Research Queueing Package (RESQ). This paper discusses the evolution of RESQ and plans for further RESQ development.

Analysis of manufacturing systems by the Research Queueing Package

Many aspects of manufacturing systems can be analyzed using simulation to model the systems behavior. The Research Queueing Package (RESQ) is a tool developed to construct and solve models of systems with jobs contending for service from many resources. The capabilities of RESQ are described in order to understand the model elements which are available for representing manufacturing systems. Then an analysis of several work-in-process (WIP) policies is presented using RESQ models solved by simulation. Four WIP management policies are analyzed and compared for a future assembly manufacturing line: (1) a push system, (2) a pull system, (3) a transfer line, and (4) a closed loop system.

The impact of distributions and disciplines on multiple processor systems

Simple queueing models are used to study the performance tradeoffs of multiple processor systems. Issues considered include the impact of CPU service disciplines and distributions, level of multiprogramming, multitasking, and job priorities.

The tree MVA algorithm

Abstract A new algorithm to solve product form queueing networks, especially those with large numbers of centers and chains, is presented. This algorithm is a Tree version of Mean Value Analysis (MVA). Tree MVA is analogous to the Tree version of Convolution developed by Lam and Lien. Like Tree Convolution, Tree MVA allows exact solution of large networks which are intractable with previous sequential algorithms. As with the sequential versions of Convolution and MVA, Tree MVA has better numerical properties than Tree Convolution. Further, Tree MVA avoids the computational complexity of sequential MVA in networks with several queue dependent centers. Thus, we consider Tree MVA to be the best algorithm for general product form networks.

RESQ: a package for solution of generalized queueing networks

RESQ (RESearch Queueing) is a tool for solution of queueing networks. The class of networks treated includes general multi-server queues, passive queues and complex routing decisions. Multiple solution techniques are provided, including numerical solution of separable balance equations and regenerative simulation. User access is provided through both interactive dialogue and a subroutine level interface.

Confidence intervals for queueing simulations of computer systems

Simulation models of computer systems may be formulated as queueing networks. Several methods for confidence interval estimation for queueing simulations are dicussed. Empirical studies of these methods are presented.

Numerical solution of some multiple chain queueing networks

Queueing networks are important as performance models of computer and communication systems because the performance of these systems is usually principally affected by contention for resources. This paper discusses exact numerical solution of certain small queueing networks which are useful in approximate numerical solution of larger networks.