CAREER: "Analysis and Control of Network Delay"

职业项目：网络延迟的分析和控制Michael J. NeelyUniversity of Southern California http：www-rcf.usc.edu/~mjneely/Abstract：这项研究调查了时变网络的通信策略，目标是为网络延迟的分析和控制开发一个基础理论。 虽然对网络吞吐量优化有很多了解，但延迟问题更复杂，也更少被理解。 延迟感知算法必须针对非常大的系统进行设计，并且必须能够最佳地适应未来网络流量、链路条件和移动的用户位置不确定的环境。 这个项目旨在建立计算严格的延迟界限的数学技术，并开发调度和路由算法，实现可证明的高吞吐量与订单最佳延迟。 这个项目还探讨了智能机制调度冗余数据包信息和利用（和控制）网络的流动性。 实际上，尽管冗余和移动性增加了决策选项的复杂性，但两者都可以有利地用于改善网络性能。本项目还介绍了数学优化的一个新课题：控制拉格朗日乘子的研究。 具体来说，众所周知，网络优化与凸规划理论密切相关，其中拉格朗日乘数可以起到类似于队列积压和/或“网络价格”的作用。“然而，在控制优化问题中使用的拉格朗日乘数的大小方面，几乎没有研究。 因此，控制拉格朗日乘数的一般理论是重要的网络延迟的研究和网络定价和经济学的相关问题。 这样的理论也可能在其他优化环境中证明是有用的，并且可能在机器学习和在线动态编程中具有更广泛的应用。

CAREER Project: Analysis and Control of Network DelayMichael J. NeelyUniversity of Southern Californiahttp://www-rcf.usc.edu/~mjneely/Abstract: This research investigates communication strategies for time-varying networks, with the goal of developing a fundamental theory for analysis and control of network delay. While much is known about network throughput optimization, the delay problem is more complex and less understood. Delay-aware algorithms must be designed for very large systems and must be able to optimally adapt to environmentswhere future network traffic, link conditions, and mobile user locations are uncertain. This project seeks to establish mathematical techniques for computing tight delay bounds, and to develop scheduling and routing algorithms that achieve provably high throughput with order-optimal delay. This project also explores intelligent mechanisms for scheduling redundant packet information and for making use of (and controlling) network mobility. Indeed, although redundancy and mobility increase the complexity of decision options, both can be used advantageously to improve network performance.This project also introduces a new topic of mathematical optimization: The study of controlling Lagrange multipliers. Specifically, it is known that network optimization is closely related to convex programming theory, where Lagrange multipliers can play a role analogous to queue backlogs and/or "network prices." However, there has been little research in the area of controlling the magnitude of the Lagrange multipliers that are used in an optimization problem. A general theory of controlled Lagrange multipliers is thus important to this study of network delay and for related problems of network pricing and economics. Such a theory may also prove useful in other optimization contexts, and may have broader applications to machine learning and online dynamic programming.