Utility Optimization Based Resource Allocation In Multipath Networks

With the popularity of networks and coming forth of new applications, the Internet has seen a great increase in traffic. Due to the limit of total transmission ca-pacity in networks, how to allocate network resources to appropriate users in reasonable manner becomes a key point to network resource allocation. In recent years, the utility theory in microeconomics is applied into the modeling and analysis of network resource allocation problems, which is regarded as a promising research point and attracts sig-nificant research attention.This dissertation considers resource allocation in multipath networks where multi-path transmission protocols are used, and presents multipath network utility maximiza-tion models. By applying the nonlinear programming theory, this dissertation analyzes the optimal resource allocation, and discusses the relationship between the prices charged by paths and those paid by users. For different objectives in resource allocation, this dissertation presents distributed rate-based flow control algorithms accordingly, and achieves the optimal resource allocation with service-oriented and user-centralized ob-jective. The main work of this dissertation can be summarized as follows:(1) The resource allocation in multipath networks is analyzed from a cross-layer point of view. Three independent sub-problems are derived from the resource allocation model for multipath networks and interpreted from an economic point of view, of which each corresponds to the objective of one layer in the Internet. Several fair resource allo-cation concepts such as proportional fairness and max-min fairness are achieved by choosing a class of utility functions, and distributed rate-based flow control algorithms are proposed, which can converge to the optimum of resource allocation model.(2) Resource allocation is also investigated in concurrent multipath networks, and for several fairness concepts, the expression of optimal resource allocation for each user is obtained. In order to obtain the optimum in a decentralized architecture, a class of distributed flow control algorithms is presented. The equilibrium point of the algorithm is identical to the optimum of resource allocation model, and is global stability based on Lyapunov stability theory for continuous dynamic systems. Thus, all trajectories along the algorithm ultimately converge to the optimal resource allocation.(3) Among the multiple concurrent paths, the pair of source and destination may only choose one as active path, leaving the remaining ones for redundancy. In this case, resource allocation on dynamic active path is modeled and analyzed. The optimal fair resource allocation for each user is obtained, and a family of distributed flow control algorithms is proposed. The equilibrium is identical to the optimum of resource alloca-tion model, and is globally stable by applying Lyapunov stability theory.(4) There are a lot of services in Internet, but can be classed into elastic and inelas-tic services based on the shapes of their utility functions. This dissertation gives the re-source allocation model for elastic services, and obtain that the prices charged by paths that a service uses are all equal. Meanwhile, the dissertation presents a class of distrib-uted flow control algorithms, which are convergent to the optimal resource allocation. Furthermore, the expression of optimal resource allocation for each service is obtained in concurrent multipath networks.(5) Since the utility functions of inelastic services are not concave, the resource al-location model for inelastic services is an intrinsically difficult problem of nonconvex optimization. For the model, the dissertation analyzes the capacities of paths used by inelastic services, and obtains that they should be provisioned with certain values such that non-negative optimal resource allocation of each service can be obtained. Based on Particle Swarm Optimization (PSO) method, the dissertation presents a nonconvex op-timization algorithm which can achieve the optimal resource allocation.(6) The dissertation considers the Universal Network architecture and proposes a mapping model for the Service Layer based on network utility maximization. Consider-ing the bandwidth threshold for inelastic services, the dissertation obtains the relation-ship between the optimal admission for an inelastic service and the optimal prices charged by links. According to mapping parameter and relationship between modules, the dissertation presents a feasible way for identifier design.The dissertation applies utility theory into resource allocation in multipath net-works, and contributes to identify the optimization objective further and provide the reasonable mechanism and algorithm in resource allocation. Furthermore, the results also have an interesting guideline for research of network architecture and designing of network protocols.