EE/CE Seminar: Insensitivity of the Mean Field of Loss Systems under Randomized SQ(d) Algorithms

Date: 3:00pm, Stepmber 15, 2017
Location: EE Conference Room (Mudd 1306) 
Speaker: Prof. Ravi R. Mazumdar, ECE Department, University of Waterloo 

Abstract: In many applications such as cloud computing, managing server farm resources etc. an incoming task or job has to be matched with an appropriate server in order to minimize the latency or blocking associated with the processing. Ideally the best choice would be to match a job to the fastest available server. However when there are thousands of servers requiring the information on all server tasks is an overkill. Pioneered in the 1990's the idea of randomized sampling of a few servers was proposed by Vvedenskaya and Dobrushin in Russia and Mitzmenmacher in the US and popularized as the "power of two" schemes which basically means that sampling two servers randomly and sending the job to the "better" server (i.e. with the shortest queue, or most resources) provides most of the benefits of sampling all the servers.

In the talk I will discuss multi-server loss models under power-of-d routing scheme when service time distributions are general with finite mean. Previous works on these models assume that the service times are exponentially distributed and insensitivity was suggested through simulations. Showing insensitivity to service time distributions has remained an open problem. The difficulty is that for general service times the underlying Markovian model is more complex. Using a measure valued process approach we first derive the mean field equation (MFE) for the empirical measure. The MFE is now characterized by a pde whose stationary point coincides with the fixed point in the case with exponential service times. This establishes the insensitivity of the fixed point. The techniques developed in this paper are applicable to study mean field limits for Markov processes on general state spaces and insensitivity properties of other queueing models.

Biography: The speaker was educated at the Indian Institute of Technology, Bombay (B.Tech, 1977), Imperial College, London (MSc, DIC, 1978) and obtained his PhD under A. V. Balakrishnan at UCLA in 1983. He is currently a University Research Chair Professor in the Dept. of ECE at the University of Waterloo, Ont., Canada where he has been since September 2004. Prior to this he was Professor of ECE at Purdue University, West Lafayette, USA. He is a D.J. Gandhi Distinguished Visiting Professor at the Indian Institute of Technology, Bombay. He has been a faculty member at Columbia (EE) and the University of Essex (Math). He is a Fellow of the IEEE and the Royal Statistical Society. He is a recipient of the Best Paper Awards at INFOCOM 2006, the International Teletraffic Congress 2015, Performance 2015, and was runner-up for the Best Paper Award at INFOCOM 1998. His research interests are in modeling, control, and performance analysis of both wireleine and wireless networks, and in applied probability and stochastic analysis with applications to queueing, statistical learning, and optimization.

Host: Prof. Javad Chaderi

500 W. 120th St., Mudd 1310, New York, NY 10027    212-854-3105               
©2014 Columbia University