Data Management Issues in Supporting Large-Scale Web Services

Data Management Issues in Supporting Large-Scale Web Services
复制标题

支持大规模 Web 服务的数据管理问题

DOI:
--
复制
发表时间:
2006
期刊:
IEEE Data Engineering Bulletin
影响因子:
--
通讯作者:
D. Shutt
D. Shutt
中科院分区:
--
文献类型:
--
作者:
P. Bernstein;Nishant Dani;Badriddine M. Khessib;Ramesh Manne;D. Shutt

文献摘要

被引文献

相似文献

本文根据MSN使用Windows Live TM服务支持大规模24×7 Web服务的经验,讨论了在支持大规模24×7 Web服务时出现的技术问题。涉及的问题包括多层体系结构、商用服务器与高级服务器的成本、管理副本、管理会话、使用物化视图以及控制检查点。我们fiNish提出了一些可能的研究方向。1介绍由服务费或广告收入支持的大型Web服务是数据库技术的主要用户。这些服务包括电子邮件、即时消息、社交网络、博客、照片分享、地图、购物和Classified。它们还包括搜索服务,尽管这些服务通常使用定制的索引引擎,而不是关系数据库系统。所谓大规模,我们指的是拥有数百万用户和数千台服务器的系统。例如,WindowsLive Messenger拥有数千台服务器和数千万个同步在线连接。它每天处理数亿次登录,每月拥有超过2.5亿独立用户。每次登录都会转换为对后端数据库服务器的查询,以进行身份验证和授权。用户每天发送数十亿条消息。消息传输本身独立于数据库而发生。据报道,Hotmail的数量也同样很大
AbstractThis paper discusses technical problems that arise in supporting large-scale 24×7 web services basedon experience at MSN with Windows Live TM services. Issues covered include multi-tier architecture,costs of commodity vs. premium servers, managing replicas, managing sessions, use of materializedviews, and controlling checkpointing. We finish with some po ssible research directions. 1 Introduction Large-scale web services supported by fee-for-service or advertising revenue are heavy users of database tech-nology. Such services include email, instant messaging, social networking, blogs, photo sharing, maps, shop-ping, and classifieds. They also include search services, th ough these typically use custom indexing engines, notrelational database systems.By large-scale, we mean systems with millions of users and thousands of servers. For example, WindowsLive Messenger has thousands of servers and tens of millions of simultaneous online connections. It processeshundreds of millions of logins per day and has over 250 million unique users per month. Each login translatesinto queries to backend database servers for authentication and authorization. Users send billions of messagesper day. The message transfer itself takes place independently of the database. Similarly large numbers havebeen reported for Hotmail