Differentiated caching services a control-theoretical approa

With the dramatic explosion of online information, the Internet is undergoing a transition from a data communication infrastructure to a global information utility. PDAs, wireless phones, webenabled vehicles, modem PCs and high-end workstations can be view

Di erentiated Caching Services; A Control-Theoretical ApproachYing Lu, Avneesh Saxena and Tarek F. Abdelzaher Department of Computer Science University of Virginia Charlottesville, VA 22903

ABSTRACTWith the dramatic explosion of online information, the Internet is undergoing a transition from a data communication infrastructure to a global information utility . PDAs, wireless phones, webenabled vehicles, modem PCs and high-end workstations can be viewed as appliances that\plug-in" to this utility for information. The increasing persity of such appliances calls for an architecture for performance di erentiation of information access. The World Wide Web is the dominant interface to the information utility today. Web proxy caching is the key performance accelerator in the web infrastructure. Client-perceived performance of Internet information access depends in great part on the local cache proxy performance. While many research e orts addressed performance di erentiation in the network, providing multiple levels of service in proxy caches has received much less attention. This paper has two main contributions. First, we describe, implement, and evaluate an architecture for di erentiated content caching services as a key element of the Internet information utility. Second, we describe a control-theoretical approach that lays well-understood theoretical foundations for resource management to achieve performance di erentiation. Digital Feedback Control Theory is leveraged to manage cache resources in a way that guarantees bounded convergence time to the speci ed performance. We describe our experiences with implementing the di erentiated caching services scheme in Squid, a popular proxy cache used by many ISPs today. Experimental study and analyses prove that di erentiated caching services provide signi cantly better performance to the premium content classes.

With the dramatic explosion of online information, the Internet is undergoing a transition from a data communication infrastructure to a global information utility. PDAs, wireless phones, webenabled vehicles, modem PCs and high-end workstations can be view

1 IntroductionThe phenomenal growth of the Internet as an information source makes world-wide information access one of its most important applications today. Internet information clients are becoming increasingly heterogeneous ranging from high-end workstations to low-end PDAs. A growing category of clients are becoming mobile, connecting to network gateways via low-bandwidth wireless links. Examples include web-enabled third-generation wireless phones, and recent vehicle models that feature a capability for wireless web access. In the near future, a much greater persi cation of information clients is envisioned. Tra c sensors, smart buildings, and various home appliances may be web-enabled, representing new data sources and sinks of the information utility. This trend for ubiquitous computing puts the web infrastructure at the center of an increasingly heterogeneous environment with di erent content types and di erent expectations regarding performance, depending on the appliance. The key performance acceleration mechanism in the web infrastructure is web caching. Caching documents

close to clients that need them reduces the number of server requests, the tra c associated with them, and the response time seen by the clients. By serving a large fraction of web requests locally, caches can signi cantly a ect end-to-end client-perceived performance of information access. Proxy cache space is not an\in nite" resource. Hence, it needs to be allocated in an e cient manner. For example, AOL reports daily tra c on their proxy caches that is in excess of 6 Terabytes of data. With a hit ratio of 60%, common to AOL caches, the cache has to fetch 6 40%= 2:4 Terabytes of new content a day. Even with secondary storage cost of as low as$5/Gigabyte, it would cost about$360,000 dollars to store a month's worth of tra c.1 This expense makes it reasonable to consider some form of QoS-aware cache resource allocation (such as service di erentiation). Current proxy caches which treat all clients and content alike might not provide the best\utility" out of their nite resources in the heterogeneous environment of the near future. To meet the increasingly persi ed performance demands of appliance-heterogeneity, in this paper we investigate an architecture for performance di erentiation in web proxy caches. The architecture allows di erent classes of content to receive di erent performance (i.e., QoS) levels. The most widely-used performance metric in the context of web caching is the cache hit ratio, H . It is related directly to performance measures of interest to the clients such as average response time. For example, if the average cache hit takes Thit time units and the average cache miss takes Tmiss time units, the average response time on information access is approximately Taccess= H Thit+(1? H )Tmiss. Assuming that Tmiss is roughly Thit+ Tbackbone, where Tbackbone is the average roundtripFault-tolerance measures such as storage redundancy in industrial-strength servers would further increase the price of the cache space.1

With the dramatic explosion of online information, the Internet is undergoing a transition from a data communication infrastructure to a global information utility. PDAs, wireless phones, webenabled vehicles, modem PCs and high-end workstations can be view

delay from the cache to an origin server, then simple algebraic manipulation yields Taccess= Tmiss? HTbackbone . If is the ratio Thit=Tbackbone, it follows that Taccess= (1+? H )Tbackbone . F …… 此处隐藏：39754字，全部文档内容请下载后查看。喜欢就下载吧 ……