Optimal threshold policies for admission control in communication networks via discrete parameter stochastic approximation

The problem of admission control of packets in communication networks is studied in the continuous time queueing framework under different classes of service and delayed information feedback. We develop and use a variant of a simulation based two timescale simultaneous perturbation stochastic approximation (SPSA) algorithm for finding an optimal feedback policy within the class of threshold type policies. Even though SPSA has originally been designed for continuous parameter optimization, its variant for the discrete parameter case is seen to work well. We give a proof of the hypothesis needed to show convergence of the algorithm on our setting along with a sketch of the convergence analysis. Extensive numerical experiments with the algorithm are illustrated for different parameter specifications. In particular, we study the effect of feedback delays on the system performance.

Neural network based optimal routing algorithm for communication networks

This paper presents the capability of the neural networks as a computational tool for solving constrained optimization problem, arising in routing algorithms for the present day communication networks. The application of neural networks in the optimum routing problem, in case of packet switched computer networks, where the goal is to minimize the average delays in the communication have been addressed. The effectiveness of neural network is shown by the results of simulation of a neural design to solve the shortest path problem. Simulation model of neural network is shown to be utilized in an optimum routing algorithm known as flow deviation algorithm. It is also shown that the model will enable the routing algorithm to be implemented in real time and also to be adaptive to changes in link costs and network topology. (C) 2002 Elsevier Science Ltd. All rights reserved.

Optimal Parameterized Policies for Resource Allocation in Communication Networks

The problem of finding optimal parameterized feedback policies for dynamic bandwidth allocation in communication networks is studied. We consider a queueing model with two queues to which traffic from different competing flows arrive. The queue length at the buffers is observed every T instants of time, on the basis of which a decision on the amount of bandwidth to be allocated to each buffer for the next T instants is made. We consider two different classes of multilevel closed-loop feedback policies for the system and use a two-timescale simultaneous perturbation stochastic approximation (SPSA) algorithm to find optimal policies within each prescribed class. We study the performance of the proposed algorithm on a numerical setting and show performance comparisons of the two optimal multilevel closedloop policies with optimal open loop policies. We observe that closed loop policies of Class B that tune parameters for both the queues and do not have the constraint that the entire bandwidth be used at each instant exhibit the best results overall as they offer greater flexibility in parameter tuning. Index Terms — Resource allocation, dynamic bandwidth allocation in communication networks, two-timescale SPSA algorithm, optimal parameterized policies. I.

An Efficient Distributed Scheme for Source Routing Protocol in Communication Networks

In this paper, we propose an efficient source routing algorithm for unicast flows, which addresses the scalability problem associated with the basic source routing technique. Simulation results indicate that the proposed algorithm indeed helps in reducing the message overhead considerably, and at the same time it gives comparable performance in terms of resource utilization across a wide range of workloads.

Location of concentrators in a computer communication network: a stochastic automation search method

The following problem is considered. Given the locations of the Central Processing Unit (ar;the terminals which have to communicate with it, to determine the number and locations of the concentrators and to assign the terminals to the concentrators in such a way that the total cost is minimized. There is alao a fixed cost associated with each concentrator. There is ail upper limit to the number of terminals which can be connected to a concentrator. The terminals can be connected directly to the CPU also In this paper it is assumed that the concentrators can bo located anywhere in the area A containing the CPU and the terminals. Then this becomes a multimodal optimization problem. In the proposed algorithm a stochastic automaton is used as a search device to locate the minimum of the multimodal cost function . The proposed algorithm involves the following. The area A containing the CPU and the terminals is divided into an arbitrary number of regions (say K). An approximate value for the number of concentrators is assumed (say m). The optimum number is determined by iteration later The m concentrators can be assigned to the K regions in (mk) ways (m > K) or (km) ways (K>m).(All possible assignments are feasible, i.e. a region can contain 0,1,…, to concentrators). Each possible assignment is assumed to represent a state of the stochastic variable structure automaton. To start with, all the states are assigned equal probabilities. At each stage of the search the automaton visits a state according to the current probability distribution. At each visit the automaton selects a 'point' inside that state with uniform probability. The cost associated with that point is calculated and the average cost of that state is updated. Then the probabilities of all the states are updated. The probabilities are taken to bo inversely proportional to the average cost of the states After a certain number of searches the search probabilities become stationary and the automaton visits a particular state again and again. Then the automaton is said to have converged to that state Then by conducting a local gradient search within that state the exact locations of the concentrators are determined This algorithm was applied to a set of test problems and the results were compared with those given by Cooper's (1964, 1967) EAC algorithm and on the average it was found that the proposed algorithm performs better.

An algorithm for dynamic optimal bandwidth allocation in communication networks

We study the problem of optimal bandwidth allocation in communication networks. We consider a queueing model with two queues to which traffic from different competing flows arrive. The queue length at the buffers is observed every T instants of time, on the basis of which a decision on the amount of bandwidth to be allocated to each buffer for the next T instants is made. We consider a class of closed-loop feedback policies for the system and use a twotimescale simultaneous perturbation stochastic approximation(SPSA) algorithm to find an optimal policy within the prescribed class. We study the performance of the proposed algorithm on a numerical setting. Our algorithm is found to exhibit good performance.

Discrete parameter simulation optimization algorithms with applications to admission control with dependent service times

We propose certain discrete parameter variants of well known simulation optimization algorithms. Two of these algorithms are based on the smoothed functional (SF) technique while two others are based on the simultaneous perturbation stochastic approximation (SPSA) method. They differ from each other in the way perturbations are obtained and also the manner in which projections and parameter updates are performed. All our algorithms use two simulations and two-timescale stochastic approximation. As an application setting, we consider the important problem of admission control of packets in communication networks under dependent service times. We consider a discrete time slotted queueing model of the system and consider two different scenarios - one where the service times have a dependence on the system state and the other where they depend on the number of arrivals in a time slot. Under our settings, the simulated objective function appears ill-behaved with multiple local minima and a unique global minimum characterized by a sharp dip in the objective function in a small region of the parameter space. We compare the performance of our algorithms on these settings and observe that the two SF algorithms show the best results overall. In fact, in many cases studied, SF algorithms converge to the global minimum.

New algorithms of the Q-learning type

We propose two algorithms for Q-learning that use the two-timescale stochastic approximation methodology. The first of these updates Q-values of all feasible state–action pairs at each instant while the second updates Q-values of states with actions chosen according to the ‘current’ randomized policy updates. A proof of convergence of the algorithms is shown. Finally, numerical experiments using the proposed algorithms on an application of routing in communication networks are presented on a few different settings.

Network flow-control using asynchronous stochastic approximation

We propose several stochastic approximation implementations for related algorithms in flow-control of communication networks. First, a discrete-time implementation of Kelly's primal flow-control algorithm is proposed. Convergence with probability 1 is shown, even in the presence of communication delays and stochastic effects seen in link congestion indications. This ensues from an analysis of the flow-control algorithm using the asynchronous stochastic approximation (ASA) framework. Two relevant enhancements are then pursued: a) an implementation of the primal algorithm using second-order information, and b) an implementation where edge-routers rectify misbehaving flows. Next, discretetime implementations of Kelly's dual algorithm and primaldual algorithm are proposed. Simulation results a) verifying the proposed algorithms and, b) comparing the stability properties are presented.

Parametrized actor-critic algorithms for finite-horizon MDPs

Due to their non-stationarity, finite-horizon Markov decision processes (FH-MDPs) have one probability transition matrix per stage. Thus the curse of dimensionality affects FH-MDPs more severely than infinite-horizon MDPs. We propose two parametrized 'actor-critic' algorithms to compute optimal policies for FH-MDPs. Both algorithms use the two-timescale stochastic approximation technique, thus simultaneously performing gradient search in the parametrized policy space (the 'actor') on a slower timescale and learning the policy gradient (the 'critic') via a faster recursion. This is in contrast to methods where critic recursions learn the cost-to-go proper. We show w.p 1 convergence to a set with the necessary condition for constrained optima. The proposed parameterization is for FHMDPs with compact action sets, although certain exceptions can be handled. Further, a third algorithm for stochastic control of stopping time processes is presented. We explain why current policy evaluation methods do not work as critic to the proposed actor recursion. Simulation results from flow-control in communication networks attest to the performance advantages of all three algorithms.

Solving MDPs using two-timescale simulated annealing with multiplicative weights

We develop extensions of the Simulated Annealing with Multiplicative Weights (SAMW) algorithm that proposed a method of solution of Finite-Horizon Markov Decision Processes (FH-MDPs). The extensions developed are in three directions: a) Use of the dynamic programming principle in the policy update step of SAMW b) A two-timescale actor-critic algorithm that uses simulated transitions alone, and c) Extending the algorithm to the infinite-horizon discounted-reward scenario. In particular, a) reduces the storage required from exponential to linear in the number of actions per stage-state pair. On the faster timescale, a 'critic' recursion performs policy evaluation while on the slower timescale an 'actor' recursion performs policy improvement using SAMW. We give a proof outlining convergence w.p. 1 and show experimental results on two settings: semiconductor fabrication and flow control in communication networks.

A reinforcement learning based algorithm for finite horizon Markov decision processes

We develop a simulation based algorithm for finite horizon Markov decision processes with finite state and finite action space. Illustrative numerical experiments with the proposed algorithm are shown for problems in flow control of communication networks and capacity switching in semiconductor fabrication.

A mathematical programming approach to optimal Markovian switching of Poisson arrival streams to queueing systems

Motivated by certain situations in manufacturing systems and communication networks, we look into the problem of maximizing the profit in a queueing system with linear reward and cost structure and having a choice of selecting the streams of Poisson arrivals according to an independent Markov chain. We view the system as a MMPP/GI/1 queue and seek to maximize the profits by optimally choosing the stationary probabilities of the modulating Markov chain. We consider two formulations of the optimization problem. The first one (which we call the PUT problem) seeks to maximize the profit per unit time whereas the second one considers the maximization of the profit per accepted customer (the PAC problem). In each of these formulations, we explore three separate problems. In the first one, the constraints come from bounding the utilization of an infinite capacity server; in the second one the constraints arise from bounding the mean queue length of the same queue; and in the third one the finite capacity of the buffer reflect as a set of constraints. In the problems bounding the utilization factor of the queue, the solutions are given by essentially linear programs, while the problems with mean queue length constraints are linear programs if the service is exponentially distributed. The problems modeling the finite capacity queue are non-convex programs for which global maxima can be found. There is a rich relationship between the solutions of the PUT and PAC problems. In particular, the PUT solutions always make the server work at a utilization factor that is no less than that of the PAC solutions.

An Actor-Critic Algorithm for Finite Horizon Markov Decision Processes

We develop a simulation based algorithm for finite horizon Markov decision processes with finite state and finite action space. Illustrative numerical experiments with the proposed algorithm are shown for problems in flow control of communication networks and capacity switching in semiconductor fabrication.

Network pricing for QoS: A 'Regulation' approach

V. S. Borkar’s work was supported in part by grant number III.5(157)/99-ET from the Department of Science and Technology, Government of India. D. Manjunath’s work was supported in part by grant number 1(1)/2004-E-Infra from the Ministry of Information Technology, Government of India.