节点文献
认知可重构的高效能Web服务体系结构研究
A Perception-based Reconfigurable Architecture for Web Service
【作者】 张帆;
【导师】 邬江兴;
【作者基本信息】 解放军信息工程大学 , 通信与信息系统, 2013, 博士
【摘要】 通常Web服务的能力依赖于通用计算平台性能的提高,但是Web服务本身的业务特征和流量规律却与通用计算机体系中以科学计算为中心的设计初衷不尽相同,因此,Web服务的性能和效能无法直接从通用计算机性能的提高中获得预期的进步。在大规模数据中心(Internet Data Center,简称IDC)应用中,随着规模和负荷动态范围的增大,Web服务的高能耗问题更加严重,已经成为高效能Web服务发展的瓶颈。为了解决这个问题,人们采取了多核处理、缓存预取、前端加速等多种研究思路,也取得了一定的理论和实践成果。这些研究成果在一定条件下获得了改进,但却未从体系结构上解决如何使服务体系与承载系统相匹配的问题。本文从体系结构着手,提出了适合Web业务特征和流量特点的架构。论文进行了如下创新性的工作:提出了基于认知的主动重构Web服务体系结构(Proactive Reconfigurable Web serviceArchitecture,简称PRWA);提出了PRWA模型的形式化定义;给出了PRWA模型的运行机理;导出了PRWA模型与现有体系结构模型之间的演化规律;理论证明了PRWA模型是一种面向Web服务的高效能模型。提出了混合可重构Web处理阵列(Hybrid Reconfigurable Web service Array,简称HRWA);给出了基于Web服务运算处理规律的算粒特征;提出了算粒提取的形式化描述方法;设计了设计了相关的原理验证系统;证明了HRWA结构是一种面向Web服务高效能计算结构。提出了面向Web服务的智能混合存储结构(Smart Hybrid Memory Architecture,简称SHMA);基于Web服务的数据结构和存储访问特征,设计了面向Web服务的智能混合存储体系;通过对词频统计和流媒体这两类典型Web服务的验证,证明了SHMA结构是一种面向Web服务的高效存储结构。设计并实现了一种基于PRWA模型的高效能Web验证平台,经过实测,验证了该平台的Web服务能效比达到普通服务器的十倍以上。
【Abstract】 The improvement of Web services mainly relies on the advance of architecture for generalcomputing. However, the characteristics of Web service do not match the architecture optimizedfor general computing. It results in a little improvements of Web service than expected.The throughput of Web service blows up over the Internet. The data center enlarges its scaleand introduces multiple layers to make service available with commodity server. Therefore itresults in a severe consequence–energy problem. Multi-core processor, prefetched cache, andfront accelerator are proposed as a part of solution to Web service. They relieve the energyproblem within a small extent. It is rare that Web service is researched from a point of view ofarchitecture. A partial acceleration cannot solve the energy trouble. It is reported that theutilization of CPU is lower than10%in data center.Our research is driven by application. Web service is I/O, storage, and computationintensive with the analysis of several typical applications. The architecture for generalcomputation cannot accommodate the diversity of Web service applications. It is the root of lowefficiency. The aim is to find an efficienty archteture for Web service. Our contrib utions can besummarized as follows:1. Proactive Reconfigurable Web service Architecture (PRWA) is proposed as a solution. Aformal model of PRWA is established. Its architecture and working mechanism are clarified. Thehigh-performance characteristics of the PRWA architecture are demonstrated. PRWA is a goodsolution to application-oriented high-performance computing and services, which has a strongtheoretical impacts and practical references.2. HRWA (Hybrid Reconfigurable Web service Array) is proposed. The computationfeatures and computing particles of Web service are summarized. A framework is put forward toextract computing particles of Web service. Then a method called BBC is designed within theframework. Futhermore, a HRWA with a set of some typ ical Web particles is implemented onFPGA to verify our idea. It is proved to be efficient for Web service.3. SHSA (smart hybrid storage architecture) is proposed based on the analysis of thecharacteristics of storage access. It is evaluated with two app lications, word count and streamingmedia service. The results show that SHSA is efficient for Web service.4. A testbed is designed for evaluation based on PRWA model. It obtains a speedup ofperformance more than ten times than that of a common Web server with our experiments.
【Key words】 High Efficiency; Web Service; Perception; Reconfigurable Technology; SmartMemory; Hybrid Memory;