Bound hierarchies for multiple-class queuing networks
Journal of the ACM (JACM) - The MIT Press scientific computation series
Accuracy, speed, and convergence of approximate mean value analysis
Performance Evaluation
A Note on the Computational Cost of the Linearizer Algorithm for Queueing Networks
IEEE Transactions on Computers
Open, Closed, and Mixed Networks of Queues with Different Classes of Customers
Journal of the ACM (JACM)
Mean-Value Analysis of Closed Multichain Queuing Networks
Journal of the ACM (JACM)
Linearizer: a heuristic algorithm for queueing network models of computing systems
Communications of the ACM
Some Extensions to Multiclass Queueing Network Analysis
Proceedings of the Third International Symposium on Modelling and Performance Evaluation of Computer Systems: Performance of Computer Systems
Numerical Methods for Unconstrained Optimization and Nonlinear Equations (Classics in Applied Mathematics, 16)
Queuing Networks with Random Selection for Service
IEEE Transactions on Software Engineering
Queuing networks with multiple closed chains: theory and computational algorithms
IBM Journal of Research and Development
Mean Value Analysis: a Personal Account
Performance Evaluation: Origins and Directions
Xaba: Exact, Approximate, and Asymptotic Solvers for Multi-class Closed Queueing Networks
TOOLS '00 Proceedings of the 11th International Conference on Computer Performance Evaluation: Modelling Techniques and Tools
| Hi-index | 0.01 |
Increasing complexity of actual computer systems has made exact modelling techniques prohibitively expensive, and the need for approximation techniques for performance evaluation is well-recognized. A new approximation technique is given for product-form queueing networks with constant-rate servers. It estimates the shift in mean queue lengths rather than the fractional deviations used in the Chandy-Neuse linearizer. Experimental results are reported which show that the new approximation 92% of the times has superior accuracy to linearizer. As population grows, the superior accuracy over linearizer increases. In 58% of the test cases, the new approximation technique gave errors of zero (at least 6 digits) while linearizer achieves such accuracy in less than 2.5% of cases. In some of the stress cases described, the new approximation technique has roughly five orders of magnitude less error than linearizer.