DC: Small:Collaborative Research: Managing Extreme-Scale Data Intensive Computing: Fundamental Design and Control Strategies

随着数据量的不断增长和计算基础设施的不断扩大，未来的极大规模数据密集型计算系统面临着前所未有的设计和控制挑战，以满足不断增长的信息处理需求。可伸缩性、健壮性、连续可用性和服务质量是确保今天设计的系统能够在未来的极端规模上以同样的效率运行所需的关键属性。本提案中描述的研究的目标是开发理论基础和实用的控制算法，使未来极端规模数据密集型计算的可扩展设计和高效管理成为可能。具体地说，研究人员将1)确定在大规模开发网络基础设施和软件系统时实现可扩展性所需的基本设计原则。这里关注的是理解由各种因素限制的性能降级，包括网络结构、处理器速度、缓冲/存储容量等。2)开发关于操作员放置、数据存储、负载减轻和资源分配的分布式控制策略，以实现高效的网内信息处理。该项目将对实现未来数据密集型信息服务系统的可伸缩性、稳健性和服务质量所需的基本设计原则和控制战略有更深入和定量的了解。这些进步将通过跨多个学科的协作努力实现，范围从性能建模、网络、排队到优化。

With the continuous growth of data production and the ever expanding computing infrastructure, future extreme-scale data intensive computing systems are facing unprecedented design and control challenges to meet the continuous and increasing demand for information processing. Scalability, robustness, continuous availability, and service quality are the key attributes desired to ensure that the system designed today is capable of operating with the same efficiency on the extreme scale of the future. The goal of the research described in this proposal is to develop theoretical foundations and practical control algorithms that enable the scalable design and efficient management of future extreme-scale data-intensive computing. Specifically, the researchers will 1) identify fundamental design principles needed to achieve scalability when developing network infrastructure and software systems in large scale. Here the concerns are to understand the performance degradation limited by various factors, including network structure, processor speeds, buffering/storage capacities, etc. 2) develop distributed control strategies on operator placement, data storage, load shedding, and resource allocation so as to enable efficient in-network information processing. The project will produce a deeper and quantitative understanding on the fundamental design principles and control strategies needed to achieve scalability, robustness and quality of service for future data-intensive information service systems. These advances will occur through a collaborative effort spanning multiple disciplines ranging from performance modeling, networking, queueing, to optimization.