Algorithmic Analysis and Congestion Control of Connection-Oriented Services in Large Scale Communication Networks.

网络技术正朝着更快、更大的网络发展，从而能够提供综合服务。这些技术发展反过来又需要分析方法的进步，以便有效地用于设计和控制未来的网络，例如基于atm的B-ISDN、全光网络和无线通信网络。新兴的高速综合通信业务要求我们分析网络系统，其中信息丢失概率范围从10**-2（如呼叫连接丢失）到10**-10（如ATM小区丢失）。在本研究中，我们将为面向连接的服务的评估和控制建立一个理论基础，这将使我们能够处理这种大范围的损失概率。首先推广了固定路由损失网络的乘积形式解。我们的公式包含了广泛的依赖于状态的呼叫生成模式、不同类型的网络服务和整个流量级别范围。利用我们在封闭排队网络模型计算算法方面的经验，我们将开发新的实用方法来估计面向连接的服务中的损失概率。为了保证近似公式的准确性，我们将制定适当的误差范围。对于具有状态依赖到达的网络，将尝试对减少负载近似方法（也称为泊松到达情况下的Erlang不动点法）进行改进和严格验证。与上述算法和渐近方法并行，我们将探索一个基于多维Ornstein-Uhlenbeck扩散过程的面向连接的网络服务模型。扩散过程方法不仅允许我们评估更一般类型的网络（即，不适合产品形式解决方案的损失/排队网络），而且还可以获得网络行为的时间依赖，即瞬态解决方案。最后，我们将制定和解决网络控制策略问题，如呼叫接纳控制和路由决策。＊＊＊

9404947 Kobayashi Network technologies have been progressing rapidly toward faster and larger networks which are capable of providing integrated services. These technical developments in turn require advances in analytic methodologies that can be effectively used in the design and control of future networks such as ATM-based B-ISDN, all-optical networks and wireless communication networks. Emerging high-speed integrated communication services require us to analyze network systems, in which information loss probabilities range from 10**-2 (e.g., call connection loss) to as small as 10**-10 (e.g., ATM cell loss). In this proposed study we will develop a theoretical foundtion for evaluation and control of connection-oriented services, which in particular will enable us to handle this wide range of loss probabilities. We begin by generalizing the product form solutions obtained for loss networks with fixed routing. Our formulation encompasses a broad range of state-dependent call generation patterns, different types of network services, and the whole range of traffic levels. Using our experience in computational algorithms for closed queueing network models, we will develop new practical methods for estimating loss probabilities in connection-oriented services. In order to insure the accuracy of approximate formulas, we will develop appropriate error bounds. A refinement and a rigorous validation of the reduced load approximation method, also referred to as the Erlang's fixed point method in the case of Poisson arrivals, will be attempted for networks with state dependent arrivals. In parallel with the above algorithmic and asymptotic methods, we will explore a model of connection-oriented network services based on a multi-dimensional Ornstein-Uhlenbeck diffusion process. The diffusion process approach will allow us not only to evaluate more general types of networks (i.e., loss/queueing networks not amenable to product form solutions) but also to obtain time-dependent, i.e., transient, solutions for network behavior. Finally, we will formulate and solve network control policy problems such as call admission control and routing decisions. ***