A Refined Mean Field Approximation

Mean field models are a popular means to approximate large and complex stochastic models that can be represented as N interacting objects. Recently it was shown that under very general conditions the steady-state expectation of any performance functional converges at rate O(1/N) to its mean field approximation. In this paper we establish a result that expresses the constant associated with this 1/N term. This constant can be computed easily as it is expressed in terms of the Jacobian and Hessian of the drift in the fixed point and the solution of a single Lyapunov equation. This allows us to propose a refined mean field approximation. By considering a variety of applications, that include coupon collector, load balancing and bin packing problems, we illustrate that the proposed refined mean field approximation is significantly more accurate that the classic mean field approximation for small and moderate values of N: relative errors are often below 1% for systems with N=10.

[1]  Diego Latella,et al.  Continuous approximation of collective system behaviour: A tutorial , 2013, Perform. Evaluation.

[2]  T. Kurtz.  Strong approximation theorems for density dependent Markov chains , 1978 .

[3]  Lei Ying,et al.  On the Approximation Error of Mean-Field Models , 2016, SIGMETRICS.

[4]  Vaidyanathan Ramaswami,et al.  Introduction to Matrix Analytic Methods in Stochastic Modeling , 1999, ASA-SIAM Series on Statistics and Applied Mathematics.

[5]  J. Eldering.  Normally Hyperbolic Invariant Manifolds: The Noncompact Case , 2013 .

[6]  Lei Ying.  Stein's Method for Mean-Field Approximations in Light and Heavy Traffic Regimes , 2017, SIGMETRICS.

[7]  Minghong Lin,et al.  Stochastic analysis of file-swarming systems , 2007, Perform. Evaluation.

[8]  R. Datko,et al.  Uniform Asymptotic Stability of Evolutionary Processes in a Banach Space , 1972 .

[9]  Nicolas Gast,et al.  Expected Values Estimated via Mean-Field Approximation are 1/N-Accurate , 2017, Proc. ACM Meas. Anal. Comput. Syst..

[10]  John N. Tsitsiklis,et al.  On the power of (even a little) centralization in distributed processing , 2011, SIGMETRICS '11.

[11]  Itai Gurvich.  Diffusion models and steady-state approximations for exponentially ergodic Markovian queues. , 2014, 1409.3393.

[12]  Moore,et al.  Unpredictability and undecidability in dynamical systems. , 1990, Physical review letters.

[13]  Bruno Gaujal,et al.  Markov chains with discontinuous drifts have differential inclusions limits , 2012 .

[14]  Yuming Jiang,et al.  On the Asymptotic Validity of the Decoupling Assumption for Analyzing 802.11 MAC Protocol , 2011, IEEE Transactions on Information Theory.

[15]  R. Srikant,et al.  Power of d Choices for Large-Scale Bin Packing , 2015, SIGMETRICS.

[16]  R. L. Dobrushin,et al.  Queueing system with selection of the shortest of two queues: an assymptotic approach , 1996 .

[17]  Richard H. Bartels,et al.  Algorithm 432 [C2]: Solution of the matrix equation AX + XB = C [F4] , 1972, Commun. ACM.

[18]  Michael Mitzenmacher,et al.  The Power of Two Choices in Randomized Load Balancing , 2001, IEEE Trans. Parallel Distributed Syst..

[19]  Benny Van Houdt.  A mean field model for a class of garbage collection algorithms in flash-based solid state drives , 2014 .

[20]  Bruno Gaujal,et al.  A mean field model of work stealing in large-scale systems , 2010, SIGMETRICS '10.

[21]  Valeria Simoncini,et al.  Computational Methods for Linear Matrix Equations , 2016, SIAM Rev..

[22]  Laurent Massoulié,et al.  Coupon replication systems , 2008, TNET.

[23]  Eric Jones,et al.  SciPy: Open Source Scientific Tools for Python , 2001 .

[24]  Vu Ngoc Phat,et al.  On the Lyapunov equation in Banach spaces and applications to control problems , 2002 .

[25]  J. Dai,et al.  Stein's method for steady-state diffusion approximations: an introduction through the Erlang-A and Erlang-C models , 2015, 1512.09364.

[26]  M. Benaïm,et al.  A class of mean field interaction models for computer and communication systems , 2008, 2008 6th International Symposium on Modeling and Optimization in Mobile, Ad Hoc, and Wireless Networks and Workshops.

[27]  Benny Van Houdt,et al.  A Fair Comparison of Pull and Push Strategies in Large Distributed Networks , 2014, IEEE/ACM Transactions on Networking.

[28]  Jean-Yves Le Boudec,et al.  A Generic Mean Field Convergence Result for Systems of Interacting Objects , 2007, Fourth International Conference on the Quantitative Evaluation of Systems (QEST 2007).

[30]  T. Kurtz.  Solutions of ordinary differential equations as limits of pure jump markov processes , 1970, Journal of Applied Probability.

[31]  J. Dai,et al.  Stein's method for steady-state diffusion approximations of $M/Ph/n+M$ systems , 2015 .

[32]  Tongxing Lu,et al.  Solution of the matrix equation AX−XB=C , 2005, Computing.