As the World Wide Web (Web) is increasingly adopted as the infrastructure for large-scale distributed information systems, issues of performance modeling become ever more critical. In particular, locality of reference is an important property in the performance modeling of distributed information systems. In the case of the Web, understanding the nature of reference locality will help improve the design of middleware, such as caching, prefetching, and document dissemination systems. For example, good measurements of reference locality would allow us to generate synthetic reference streams with accurate performance characteristics, would allow us to compare empirically measured streams to explain differences, and would allow us to predict expected performance for system design and capacity planning. In this paper we propose models for both temporal and spatial locality of reference in streams of requests arriving at Web servers. We show that simple models based only on document popularity (likelihood of reference) are insufficient for capturing either temporal or spatial locality. Instead, we rely on an equivalent, but numerical, representation of a reference stream: a stack distance trace. We show that temporal locality can be characterized by

Prefetching and caching are techniques commonly used in I/O systems to reduce latency. Many researchers have advocated the use of caching and prefetching to reduce latency in the Web. We derive several bounds on the performance improvements seen from these techniques, and then use traces of Web proxy activity taken at Digital Equipment Corporation to quantify these bounds. We found that for these traces, local proxy caching could reduce latency by at best 26%, prefetching could reduce latency by at best 57%, and a combined caching and prefetching proxy could provide at best a 60% latency reduction. Furthermore, we found that how far in advance a prefetching algorithm was able to prefetch an object was a significant factor in its ability to reduce latency. We note that the latency reduction from caching is significantly limited by the rapid changes of objects in the Web. We conclude that for the workload studied caching offers moderate assistance in reducing latency. Prefetching can of...

We present two server-initiated protocols to improve the performance of distributed information systems (e.g. WWW). Our first protocol is a hierarchical data dissemination mechanism that allows information to propagate from its producers to servers that are closer to its consumers. This dissemination reduces network traffic and balances load amongst servers by exploiting geographic and temporal locality of reference properties exhibited in client access patterns. Our second protocol relies on "speculative service", whereby a request for a document is serviced by sending, in addition to the document requested, a number of other documents that the server speculates will be requested inthenear future. This speculation reduces service time by exploiting the spatial locality of reference property. We present results of trace-driven simulations that quantify the attainable performance gains for both protocols.

Speculative service implies that a client's request for a document is serviced by sending, in addition to the document requested, a number of other documents (or pointers thereto) that the server speculates will be requested by the client in the near future. This speculation is based on statistical information that the server maintains for each document it serves. The notion of speculative service is analogous to prefetching, which is used to improve cache performance in distributed /parallel shared memory systems, with the exception that servers (not clients) control when and what to prefetch. Using trace simulations based on the logs of our departmental HTTP server http://cs-www.bu.edu,weshow that both server load and service time could be reduced considerably, if speculative service is used. This is above and beyond what is currently achievable using client-side caching [3] and server-side dissemination [2]. We identify a number of parameters that could be used to fine-tune the level of speculation performed by the server.

. We study Markov models whose state spaces arise from the Cartesian product of two or more discrete random variables. We show how to parameterize the transition matrices of these models as a convex combination---or mixture---of simpler dynamical models. The parameters in these models admit a simple probabilistic interpretation and can be fitted iteratively by an Expectation-Maximization (EM) procedure. We derive a set of generalized Baum-Welch updates for factorial hidden Markov models that make use of this parameterization. We also describe a simple iterative procedure for approximately computing the statistics of the hidden states. Throughout, we give examples where mixed memory models provide a useful representation of complex stochastic processes. Keywords: Markov models, mixture models, discrete time series 1. Introduction The modeling of time series is a fundamental problem in machine learning, with widespread applications. These include speech recognition (Rabiner, 1989), natu...

World-Wide Web (WWW) services have grown to levels where signi cant delays are expected to happen. Techniques like pre-fetching are likely to help users to personalize their needs, reducing their waiting times. However, prefetching is only e ective if the right documents are identi ed and if user's move is correctly predicted. Otherwise, pre-fetching will only waste bandwidth. Therefore, it is productive to determine whether a revisit will occur or not, before starting pre-fetching. In this paper we develop two user models that help determining user's next move. One model uses Random Walk approximation and the other is based on Digital Signal Processing techniques. We also give hints on how to use such models with a simple pre-fetching technique that we are developing. 1

HTTP accounts for most of the bytes flowing over the Internet backbone (up to 75 percent, in one study [1]). This bandwidth demand requires continued investment in link and switch capacity, and leads to congestion, which increases user-perceived latency. At the edges of the Internet, which are often bandwidth-constrained, every extra byte transferred adds incremental delay; this is a particular problem for home users, most of whom do not yet have a cost-effective means to increase bandwidth above 56 kb/s. And every round-trip through the Internet adds delay, often several hundred milliseconds. Almost any computer system that suffers from latency or bandwidth problems can benefit from caching. The Web is no exception, and caching mechanisms have been part of HTTP almost since its inception. Caching is perhaps the one aspect of the Web most easily amenable to academic studies, and many research papers have been published. Web caching is also sufficiently useful to have led to the creation of a rapidly growing industry. Caching works when a reference stream has locality. Temporal locality exists when an item is referenced more than once — a cache can store the item on the first reference, and then return it for subsequent references. Traditionally, Web caches have exploited temporal locality, with a URL as the granularity of reference. Such a cache stores a response to a request for a URL, and then a subsequent request for the same URL yields a cache hit. Other caches in computer systems, such as CPU data caches, often reach hit rates approaching 100 percent, but numerous studies of actual Web reference streams report much lower hit rates, often 50 percent or less. Indeed, recent studies have shown intrinsic limits to the hit rates achievable with URL-granularity temporal locality: no matter how large the cache or user population, many references will never be cache hits. This article surveys some of the techniques proposed to

Many HTTP resources (pages, graphics, etc.) are exact duplicates of other resources with different URLs. If an HTTP cache contains a duplicate of a requested resource, and could detect this, it could avoid substantial network costs by returning the cached duplicate in place of the requested URL. Previous studies have shown that there is substantial duplication of content in both HTTP and FTP, and several protocols have been proposed to support efficient and safe duplicate suppression in HTTP. We use traces covering millions of HTTP requests to quantify the potential benefit of an HTTP duplicate-suppression extension. In particular, we show that the benefits vary depending on content-type, and that a small fraction of Web servers account for most of the duplicated resources.

Most modern cache systems treat each access as an independent event. Events in a computer system, however are driven by programs. Thus, accesses occur in distinct sequences and are by no means random. The result is that modern caches ignore useful information. In a Unix le system, when the executable for the program make is referenced, it is likely that references to les such ascc, as, and ld will soon follow. To successfully track the le reference patterns, a model must e ciently handle the large number of distinct les, as well as the constant addition and removal of les. To further complicate matters, the permanent nature of a le system requires that this model be able to run continuously, forcing any realistic model to address issues of long-term model buildup and adapting to changing patterns. We have adapted a multi-order context modeling technique used in the data com-pression method Prediction by Partial Match (PPM) to track le reference patterns. We then modi ed this context modeling technique to e ciently handle large numbers of distinct les, and to adapt to areas of local activity. From this model, we are able to determine

Prefetching and Web caching have been known as techniques to increase the speed of Web loading process. Previous studies have been conducted to infuse artificial intelligence such as Artificial Neural Network (ANN) into Web caching. In this paper, we propose a new hybrid technique based on combination of ANN and Particle Swarm Optimization (PSO) for classification Web object either to cache or not and generate rules from log data by using Rough Set technique on proxy server (Rough Neuro-PSO). It is needed because mobile context has limited resources like speed and memory. Our method is by using XML file for prefetching which is saved into mobile memory. Prefetching that used xml file is much faster to be searched or accessed. In Web caching side, we enhance the speed by using Rough Neuro-PSO in order to choose the log. At the end of this paper, we present a new framework that is believed to speed up the access of Web page in mobile environment context. 1.