Traffic management and design of multiservice networks: The Internet and ATM networks.

摘要

This work starts by considering flow control mechanisms for rate-adaptive services in networks with static connections. It spans performance and design of dynamic networks supporting rate-adaptive services, and culminates in a collection of tools and methods for designing multiservice networks. These results show some guidelines to the traffic management and design of networks for both network designer/provider and user benefits.; We consider a flow control algorithm to allocate bandwidth for rate-adaptive services in a network with ‘fixed’ number of connections subject to throughput and fairness constraints. Our algorithm achieves a max-min fair rate allocation among contending users, and has desirable properties in that it can operate in a decentralized and asynchronous manner The algorithm is simple in that the network links make local measurements of capacity and calculate local ‘explicit rates’ without requiring knowledge of the number of ongoing connections. Connections will receive bandwidth determined by the minimum explicit rate along their respective routes. We discuss its stability, convergence, and feasibility issues related to fair allocation and rate-based flow control. We also consider the role of sessions with priorities under weighted max-min fair allocation of bandwidth, and its use for ‘ABR flow control’ in ATM networks.; We next consider the stability and performance of a model for ‘dynamic’ networks supporting rate-adaptive services. In our model connection arrivals are stochastic and have a random amount of data to send, so the number of connections in the system changes over time. In turn bandwidth allocated to connections also may change over time due to feedback control, e.g ., max-min fair or proportionally fair allocation of bandwidth, that reacts to congestion and implicitly to the number of ongoing connections. We prove the stability of such networks when offered load on each link does not exceed its capacity. Simulations are used to investigate the performance, in terms of average connection delays, for various types of bandwidth allocation. In addition bounds on the performance are derived. Our model can be used to investigate connection level stability and performance of networks supporting rate-adaptive services. We also discuss design issues and possible methods to guarantee delay quality of service requirements to dynamic connections, as required by interactive services.; We then consider multiservice ATM networks, in which rate-adaptive ABR and CBR services, with dynamic arrivals and departures, are sharing a single node. This is modeled by two-dimensional Markov chain, and a matrix-geometric equation is solved to yield performance estimates for ABR connections, i.e., average delay and available bandwidth. By a “separation of time scales” between ABR and CBR services, we propose an approximate solution for the steady state performance of the above Markov chain. These performance results enable joint design of networks supporting multiple services. These results are extended to large-scale networks to compute available bandwidth for ABR connections in a dynamically changing environment. We find an upper bound on the average minimum throughput for ABR services and show that the bound is asymptotically achieved in large-capacity networks. To further increase efficiency, we consider adjustments via network level priority by way of weighted max-min fair allocation of bandwidth.