CAREER: Algorithm Design for Optimization Problems in Network Over-Provisioning

互联网应用正在以惊人的速度增长。它们的范围从个人到商业，再到关键任务。所有这些都要求网络服务的可靠性、高性能和安全性。这些需求的理由主要是由于网络技术和资源可用性方面的最新进步。网络带宽、信息存储、计算机处理能力等互联网资源不再是过去的瓶颈。然而，不断增长的流量需求、流量的动态变化以及恶意用户的存在对满足这些需求构成了严峻的挑战。今天的互联网还没有完全克服这些挑战，本项目研究了为实现可靠性、高性能和安全性而对重要类别的网络系统进行定量资源过度配置的问题。简而言之，超额拨备的原则是根据预期需求保守地分配资源。资源过度调配可实现更好的不间断服务，并更好地管理流量动态和攻击。但是，资源过度调配不能以临时方式进行。粗心大意地过度配置资源会增加金钱成本、维护开销，在某些情况下还会降低性能和安全性。因此，重要的是要在数量上超额提供资源，以避免这些问题。要进行定量的资源过度配置，需要解决一系列问题，包括要配置多少额外资源，以及如何以最佳方式配置额外资源。本项目的重点是资源过度配置中此类优化问题的算法设计。该项目对网络领域产生了直接而重大的影响。量化的资源过度配置可以作为有效地向互联网应用提供高水平的可靠性、性能和安全性的服务的重要基础。

Internet applications are growing at a phenomenal rate. They range from personal to commercial to mission critical. All of them demand network service reliability, high performance and security. The justification for these demands is mainly due to recent advances in network technologies and resource availability. Internet resources such as network bandwidth, information storage, and computer processing power are no longer the bottlenecks that used to be so in the past. However, the ever increasing traffic demand, dynamics in traffic, and the presence of malicious users pose serious challenges towards meeting these demands. Today's Internet has yet fully overcome these challenges.This project studies quantitative resource over-provisioning for important classes of network systems to achieve reliability, high performance and security. In simple terms, the over-provisioning principle is one where resources are allocated conservatively based on expected demands. Resource over-provisioning enables better uninterrupted services and better management of traffic dynamics and attacks. However, resource over-provisioning cannot be done in an ad hoc manner. Careless over-provisioning of resources comes with increased monetary costs, maintenance overhead, and in some cases performance andsecurity degradation as well. Hence, it is important that resources are quantitatively over-provisioned to avoid these problems. To conduct quantitative resource over- provisioning, a host of issues need to be resolved including how many extra resources to provision, and how to optimally provision the extra resources. The focus of this project is the design of algorithms for such optimization problems in resource over-provisioning. This project has immediate and significant impacts on the networking arena. Quantitative resource over-provisioning can serve as an important foundation to effectively provide services with high levels of reliability, performance and security to internet applications.