Modeling and performance analysis of distributed computer networks with differentiated services.

摘要

A network with differentiated services guarantees the provision of Quality of service (QoS) for different traffic classes by incorporating various class-based scheduling disciplines. Recently, there has been interest in the study of network performance modeling in networks with differentiated services. In particular, the performance modeling of differentiated service networks in distributed architectures has attracted several network communities. The heterogeneity of user traffic can have major impact on the overall performance of resource allocation problems in computer networks. This work aims at shedding light on the performance modeling of distributed wired and wireless environments with differentiated services. We study the performance of priority-based distributed schemes for several important network applications: (i) network routing (load balancing) in wired network (ii) network selection in wireless environment (iii) scheduling in Body Area Networks and (vi) Interference mitigation scheme for Wireless Body Area Networks. We will utilize Game-theoretic models to model the interaction between network entities and evaluate the system performance analytically and empirically. Our main objective is to investigate the degree in which various service differentiation mechanisms have impact on the overall performance of distributed networks. We analyze several important equilibria properties such as existence, convergence time, and inefficiency for the studied network scenarios.;First, we study a network of parallel links where each link incorporates General Processor Sharing (GPS) scheduling mechanism. Using the non-cooperative game model, we investigate the existence and uniqueness of equilibrium points. We study the efficiency of equilibria by deriving an upper bound on the Price of Anarchy (PoA). Finally, we utilize the simulation to compare our PoA bound with actual bound.;Second, we study wireless network selection in a distributed environment with Discriminatory Processor Sharing (DPS) scheduling discipline. We consider a set of wireless clients in a network composed of multiple base stations with differentiated services. The goal of each wireless client is to assign itself to a wireless base station that gives the higher throughput. We formulate the interaction between clients as a weighted congestion game and investigate the existence of equilibria. We give several regimes under which the equilibrium is guaranteed to exist. These regimes are implemented either on the base stations or wireless clients. Furthermore, we derive several upper bounds on the equilibrium convergence time and give an extensive simulation to evaluate our results.;Third, we consider the interaction between autonomous Wireless Body Area Networks (WBAN) in a crowded environment such as hospital. We propose a distributed approach that mitigates the inter-WBAN interference using the game theory models. The interaction between WBANs is modeled as an infinite repeated game. We propose a "grim" strategy with Subgame Perfect Equilibrium (SPE) property under specific conditions. Furthermore, we investigate the equilibrium by comparing that to a sub-optimal Pareto strategy. We test our results using a specific simulator designed for WBAN.;Finally, we propose a new QoS framework for WBAN architecture. The goal of this framework is to implement a separate QoS layer in the design of WBAN stack protocol so that it becomes independent of the application layer. The framework offers an application profile interface where each application can submit its QoS requirements.