Abstract
We introduce AutoCAT, a class of algorithms to automatically generate exact and approximate product-form solutions for large Markov processes that cannot be solved by direct numerical methods. Focusing on models that can be described as cooperating Markov processes, which include queueing networks and stochastic Petri nets as special cases, it is shown that finding a global optimum for a nonconvex quadratic program is sufficient for determining a product-form solution. Such problems are notoriously hard to solve due to the inherent difficulty of searching over nonconvex sets. Using a potential theory for Markov processes, convexification techniques, and a class of linear constraints that follow from stochastic characterization of product-form solutions, we obtain a family of linear programming relaxations that can be solved efficiently. A sequence of these linear programs is solved under increasingly tighter constraints to determine the exact product-form solution of a model when one exists. This approach is then extended to obtain approximate solutions for non-product-form models. Finally, our new techniques are validated with examples and increasingly complex case studies that show the effectiveness of the method on both conventional and novel performance models.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
Note that all test cases did converge, but no rigorous convergence proof is available.
References
Al-Khayyal, F.A., Falk, J.E.: Jointly constrained biconvex programming. Math. Oper. Res. 8(2), 273–286 (1983)
Argent-Katwala, A.: Automated product-forms with Meercat. In: Proceedings of SMCTOOLS, October 2006
Balbo, G., Bruell, S.C., Sereno, M.: Product form solution for generalized stochastic Petri nets. IEEE TSE 28(10), 915–932 (2002)
Balsamo, S., Onvural, R.O., De Nitto Personé, V.: Analysis of Queueing Networks with Blocking. Kluwer, Norwell, MA (2001)
Balsamo, S., Dei Rossi, G., Marin, A.: A numerical algorithm for the solution of product-form models with infinite state spaces. In Computer Performance Engineering (A. Aldini, M. Bernardo, L. Bononi, V. Cortellessa, Eds.) LNCS 6342, Springer 2010. (7th Europ. Performance Engineering Workshop EPEW 2010, Bertinoro (Fc), Italy, (2010)
Baskett, F., Chandy, K.M., Muntz, R.R., Palacios, F.G.: Open, closed, and mixed networks of queues with different classes of customers. J. ACM 22(2), 248–260 (1975)
Bertsimas, D., Tsitsiklis, J.: Introduction to Linear Optimization. Athena Scientific, Nashua, NH (1997)
Bolch, G., Greiner, S., de Meer, H., Trivedi, K.S.: Queueing Networks and Markov Chains. Wiley, New York (1998)
Buchholz, P.: Product form approximations for communicating Markov processes. In: Proceedings of QEST, pp. 135–144. IEEE, New York (2008)
Buchholz, P.: Product form approximations for communicating Markov processes. Perform. Eval. 67(9), 797–815 (2010)
Burer, S., Letchford, A.N.: On nonconvex quadratic programming with box constraints. SIAM J. Optim. 20(2), 1073–1089 (2009)
Cao, X.R.: The relations among potentials, perturbation analysis, and Markov decision processes. Discr. Event Dyn. Sys. 8(1), 71–87 (1998)
Cao, X.R., Ma, D.J.: Performance sensitivity formulae, algorithms and estimates for closed queueing networks with exponential servers. Perform. Eval. 26, 181–199 (1996)
Casale, G., Harrison, P.G.: A class of tractable models for run-time performance evaluation. In: Proceedings of ACM/SPEC ICPE (2012)
Casale, G., Harrison, P.G., Vigliotti, M.G.: Product-form approximation of queueing networks with phase-type service. ACM Perf. Eval. Rev. 39(4) (2012)
de Souza e Silva, E., Ochoa, P.M.: State space exploration in Markov models. In: Proceedings of ACM SIGMETRICS, pp. 152–166 (1992)
Dijk, N.: Queueing Networks and Product Forms: A Systems Approach. Wiley, Chichester (1993)
Fourneau, J.M., Quessette, F.: Computing the steady-state distribution of G-networks with synchronized partial flushing. In: Proceedings of ISCIS, pp. 887–896. Springer, Berlin (2006)
Fourneau, J.M., Plateau, B., Stewart, W.: Product form for stochastic automata networks. In: Proceedings of ValueTools, pp. 1–10 (2007)
Gelenbe, E.: Product-form queueing networks with negative and positive customers. J. App. Probab. 28(3), 656–663 (1991)
GNU GLPK 4.8. http://www.gnu.org/software/glpk/
Harrison, P.G.: Turning back time in Markovian process algebra. Theor. Comput. Sci 290(3), 1947–1986 (2003)
Harrison, P.G.: Reversed processes, product forms and a non-product form. Lin. Algebra Appl. 386, 359–381 (2004)
Harrison, P.G., Hillston, J.: Exploiting quasi-reversible structures in Markovian process algebra models. Comp. J. 38(7), 510–520 (1995)
Harrison, P.G., Lee, T.: Separable equilibrium state probabilities via time reversal in Markovian process algebra. Theor. Comput. Sci 346, 161–182 (2005)
Harrison, P.G., Llado, C.: A PMIF with Petri net building blocks. In: Proceedings of ICPE (2011)
Hillston, J.: A compositional approach to performance modelling. Ph.D. Thesis, University of Edinburgh (1994)
IBM ILOG CPLEX 12.0 User’s Manual, 2010
Jackson, J.R.J.: Jobshop-like queueing systems. Manage. Sci. 10(1), 131–142 (1963)
Kelly, F.P.: Networks of queues with customers of different types. J. Appl. Probab. 12(3), 542–554 (1975)
Kelly, F.P.: Reversibility and Stochastic Networks. Wiley, New York (1979)
Liu, Z.: Performance analysis of stochastic timed Petri nets using linear programming approach. IEEE TSE 11(24), 1014–1030 (1998)
Löfberg, J.: YALMIP: A toolbox for modeling and optimization in MATLAB. In: Proceedings of CACSD (2004)
Marin, A., Bulò, S.R.: A general algorithm to compute the steady-state solution of product-form cooperating Markov chains. In: Proceedings of MASCOTS, pp. 1–10 (2009)
Marin, A., Vigliotti, M.G.: A general result for deriving product-form solutions in Markovian models. In: Proceedings of ICPE, pp. 165–176 (2010)
McCormick, G.P.: Computability of global solutions to factorable nonconvex programs. Math. Prog. 10, 146–175 (1976)
Muntz, R.R.: Poisson departure processes and queueing networks. Tech. Rep. RC 4145, IBM T.J. Watson Research Center, Yorktown Heights, NY (1972)
Nelson, R.D.: The mathematics of product form queuing networks. ACM Comp. Surv. 25(3), 339–369 (1993)
Nocedal, J., Wright, S.J.: Numerical Optimization. Springer, Berlin (1999)
Plateau, B.: On the stochastic structure of parallelism and synchronization models for distributed algorithms. SIGMETRICS 147–154 (1985)
Saad, Y.: Iterative Methods for Sparse Linear Systems. SIAM, Philadelphia (2000)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Appendix
Infinite Processes
Numerical optimization techniques generally require matrices of finite size. In both ENS and its relaxations, we therefore used exact or approximate aggregations to truncate the state spaces of any infinite processes. Let C + 1 be the maximum acceptable matrix order. Then we decompose the generator matrix and its equilibrium probability vector of an infinite process k as
Finally, we comment on the choice of the parameter C for a given process k. Since this determines the number of states N k for the truncated process, an optimal choice of this value can provide substantial computational savings. Let us first note that starting from a small C, it is easy to integrate additional constraints or potential vectors in the linear formulations for a value C ′ > C. Recall that we propose in the rest of the paper a sequence of linear programs in order to obtain a feasible solution x. Then, if a linear program is infeasible, this can be due either to a lack of a product form or to a truncation where C is too small. The latter case can be readily diagnosed by adding slack variables, as in QCP, to the ergodicity condition and verifying if this is sufficient to restore feasibility. In such a case, the C value is updated to the smallest value such that feasibility is restored in the main linear program.
ZPR Example Model
Structural Approximation Pseudocode
G-Network Case Study
We report the parameters for the G-network given in [5]. The network consists of M = 10 queues with exponentially distributed service times having rates μ1 = 4. 5 and \({\mu }_{i} = 4.0 + (0.1)i\) for i ∈ [2, 10]. The external arrival rate defines a Poisson process with rate λ = 5. 0. The routing matrix for (positive) customers has in row i and column j the probability r i, j  +  of a (positive) customer being routed to queue j, as a positive customer, upon leaving queue i. In this case study, this routing matrix is given by
Loss and BBS Models
The model is composed of M = 5 queues that cooperate on a set of A = 12 actions, one for each possible job movement from and inside the network. The routing probabilities R[k, j] from queue k to queue j are as follows:
Rights and permissions
Copyright information
© 2013 Springer Science+Business Media New York
About this paper
Cite this paper
Casale, G., Harrison, P.G. (2013). AutoCAT: Automated Product-Form Solution of Stochastic Models. In: Latouche, G., Ramaswami, V., Sethuraman, J., Sigman, K., Squillante, M., D. Yao, D. (eds) Matrix-Analytic Methods in Stochastic Models. Springer Proceedings in Mathematics & Statistics, vol 27. Springer, New York, NY. https://doi.org/10.1007/978-1-4614-4909-6_4
Download citation
DOI: https://doi.org/10.1007/978-1-4614-4909-6_4
Published:
Publisher Name: Springer, New York, NY
Print ISBN: 978-1-4614-4908-9
Online ISBN: 978-1-4614-4909-6
eBook Packages: Mathematics and StatisticsMathematics and Statistics (R0)