Measuring the Behavior of a World-Wide Web Server

Server performance has become a crucial issue for improving the overall performance of the World-Wide Web. This paper describes Webmonitor, a tool for evaluating and understanding server performance, and presents new results for a realistic workload. Webmonitor measures activity and resource consumption, both within the kernel and in HTTP processes running in user space. Webmonitor is implemented using an efficient combination of sampling and event-driven techniques that exhibit low overhead. Our initial implementation is for the Apache World-Wide Web server running on the Linux operating system. We demonstrate the utility of Webmonitor by measuring and understanding the performance of a Pentium-based PC acting as a dedicated WWW server. Our workload uses a file size distribution with a heavy tail. This captures the fact that Web servers must concurrently handle some requests for large audio and video files, and a large number of requests for small documents, containing text or images. Our results show that in a Web server saturated by client requests, over 90% of the time spent handling HTTP requests is spent in the kernel. Furthermore, keeping TCP connections open, as required by TCP, causes a factor of 2-9 increase in the elapsed time required to service an HTTP request. Data gathered from Webmonitor provide insight into the causes of this performance penalty. Specifically, we observe a significant increase in resource consumption along three dimensions: the number of HTTP processes running at the same time, CPU utilization, and memory utilization. These results emphasize the important role of operating system and network protocol implementation in determining Web server performance.

ImageRover: A Content-Based Image Browser for the World Wide Web

ImageRover is a search by image content navigation tool for the world wide web. To gather images expediently, the image collection subsystem utilizes a distributed fleet of WWW robots running on different computers. The image robots gather information about the images they find, computing the appropriate image decompositions and indices, and store this extracted information in vector form for searches based on image content. At search time, users can iteratively guide the search through the selection of relevant examples. Search performance is made efficient through the use of an approximate, optimized k-d tree algorithm. The system employs a novel relevance feedback algorithm that selects the distance metrics appropriate for a particular query.

Generating Representative Web Workloads for Network and Server Performance Evaluation

One role for workload generation is as a means for understanding how servers and networks respond to variation in load. This enables management and capacity planning based on current and projected usage. This paper applies a number of observations of Web server usage to create a realistic Web workload generation tool which mimics a set of real users accessing a server. The tool, called Surge (Scalable URL Reference Generator) generates references matching empirical measurements of 1) server file size distribution; 2) request size distribution; 3) relative file popularity; 4) embedded file references; 5) temporal locality of reference; and 6) idle periods of individual users. This paper reviews the essential elements required in the generation of a representative Web workload. It also addresses the technical challenges to satisfying this large set of simultaneous constraints on the properties of the reference stream, the solutions we adopted, and their associated accuracy. Finally, we present evidence that Surge exercises servers in a manner significantly different from other Web server benchmarks.

On the Interaction Between an Operating System and Web Server

This paper examines how and why web server performance changes as the workload at the server varies. We measure the performance of a PC acting as a standalone web server, running Apache on top of Linux. We use two important tools to understand what aspects of software architecture and implementation determine performance at the server. The first is a tool that we developed, called WebMonitor, which measures activity and resource consumption, both in the operating system and in the web server. The second is the kernel profiling facility distributed as part of Linux. We vary the workload at the server along two important dimensions: the number of clients concurrently accessing the server, and the size of the documents stored on the server. Our results quantify and show how more clients and larger files stress the web server and operating system in different and surprising ways. Our results also show the importance of fixed costs (i.e., opening and closing TCP connections, and updating the server log) in determining web server performance.

Basis Token Consistency: A Practical Mechanism for Strong Web Cache Consistency

With web caching and cache-related services like CDNs and edge services playing an increasingly significant role in the modern internet, the problem of the weak consistency and coherence provisions in current web protocols is becoming increasingly significant and drawing the attention of the standards community [LCD01]. Toward this end, we present definitions of consistency and coherence for web-like environments, that is, distributed client-server information systems where the semantics of interactions with resource are more general than the read/write operations found in memory hierarchies and distributed file systems. We then present a brief review of proposed mechanisms which strengthen the consistency of caches in the web, focusing upon their conceptual contributions and their weaknesses in real-world practice. These insights motivate a new mechanism, which we call "Basis Token Consistency" or BTC; when implemented at the server, this mechanism allows any client (independent of the presence and conformity of any intermediaries) to maintain a self-consistent view of the server's state. This is accomplished by annotating responses with additional per-resource application information which allows client caches to recognize the obsolescence of currently cached entities and identify responses from other caches which are already stale in light of what has already been seen. The mechanism requires no deviation from the existing client-server communication model, and does not require servers to maintain any additional per-client state. We discuss how our mechanism could be integrated into a fragment-assembling Content Management System (CMS), and present a simulation-driven performance comparison between the BTC algorithm and the use of the Time-To-Live (TTL) heuristic.

Differentiated Control of Web Traffic: A Numerical Analysis

Internet measurements show that the size distribution of Web-based transactions is usually very skewed; a few large requests constitute most of the total traffic. Motivated by the advantages of scheduling algorithms which favor short jobs, we propose to perform differentiated control over Web-based transactions to give preferential service to short web requests. The control is realized through service semantics provided by Internet Traffic Managers, a Diffserv-like architecture. To evaluate the performance of such a control system, it is necessary to have a fast but accurate analytical method. To this end, we model the Internet as a time-shared system and propose a numerical approach which utilizes Kleinrock's conservation law to solve the model. The numerical results are shown to match well those obtained by packet-level simulation, which runs orders of magnitude slower than our numerical method.

Safe Composition of Web Communication Protocols for Extensible Edge Services

As new multi-party edge services are deployed on the Internet, application-layer protocols with complex communication models and event dependencies are increasingly being specified and adopted. To ensure that such protocols (and compositions thereof with existing protocols) do not result in undesirable behaviors (e.g., livelocks) there needs to be a methodology for the automated checking of the "safety" of these protocols. In this paper, we present ingredients of such a methodology. Specifically, we show how SPIN, a tool from the formal systems verification community, can be used to quickly identify problematic behaviors of application-layer protocols with non-trivial communication models—such as HTTP with the addition of the "100 Continue" mechanism. As a case study, we examine several versions of the specification for the Continue mechanism; our experiments mechanically uncovered multi-version interoperability problems, including some which motivated revisions of HTTP/1.1 and some which persist even with the current version of the protocol. One such problem resembles a classic degradation-of-service attack, but can arise between well-meaning peers. We also discuss how the methods we employ can be used to make explicit the requirements for hardening a protocol's implementation against potentially malicious peers, and for verifying an implementation's interoperability with the full range of allowable peer behaviors.

On the Intrinsic Locality Properties of Web Reference Streams

There has been considerable work done in the study of Web reference streams: sequences of requests for Web objects. In particular, many studies have looked at the locality properties of such streams, because of the impact of locality on the design and performance of caching and prefetching systems. However, a general framework for understanding why reference streams exhibit given locality properties has not yet emerged. In this work we take a first step in this direction, based on viewing the Web as a set of reference streams that are transformed by Web components (clients, servers, and intermediaries). We propose a graph-based framework for describing this collection of streams and components. We identify three basic stream transformations that occur at nodes of the graph: aggregation, disaggregation and filtering, and we show how these transformations can be used to abstract the effects of different Web components on their associated reference streams. This view allows a structured approach to the analysis of why reference streams show given properties at different points in the Web. Applying this approach to the study of locality requires good metrics for locality. These metrics must meet three criteria: 1) they must accurately capture temporal locality; 2) they must be independent of trace artifacts such as trace length; and 3) they must not involve manual procedures or model-based assumptions. We describe two metrics meeting these criteria that each capture a different kind of temporal locality in reference streams. The popularity component of temporal locality is captured by entropy, while the correlation component is captured by interreference coefficient of variation. We argue that these metrics are more natural and more useful than previously proposed metrics for temporal locality. We use this framework to analyze a diverse set of Web reference traces. We find that this framework can shed light on how and why locality properties vary across different locations in the Web topology. For example, we find that filtering and aggregation have opposing effects on the popularity component of the temporal locality, which helps to explain why multilevel caching can be effective in the Web. Furthermore, we find that all transformations tend to diminish the correlation component of temporal locality, which has implications for the utility of different cache replacement policies at different points in the Web.

Combining Textual and Visual Cues for Content-based Image Retrieval on the World Wide Web

Some WWW image engines allow the user to form a query in terms of text keywords. To build the image index, keywords are extracted heuristically from HTML documents containing each image, and/or from the image URL and file headers. Unfortunately, text-based image engines have merely retro-fitted standard SQL database query methods, and it is difficult to include images cues within such a framework. On the other hand, visual statistics (e.g., color histograms) are often insufficient for helping users find desired images in a vast WWW index. By truly unifying textual and visual statistics, one would expect to get better results than either used separately. In this paper, we propose an approach that allows the combination of visual statistics with textual statistics in the vector space representation commonly used in query by image content systems. Text statistics are captured in vector form using latent semantic indexing (LSI). The LSI index for an HTML document is then associated with each of the images contained therein. Visual statistics (e.g., color, orientedness) are also computed for each image. The LSI and visual statistic vectors are then combined into a single index vector that can be used for content-based search of the resulting image database. By using an integrated approach, we are able to take advantage of possible statistical couplings between the topic of the document (latent semantic content) and the contents of images (visual statistics). This allows improved performance in conducting content-based search. This approach has been implemented in a WWW image search engine prototype.

Changes in Web Client Access Patterns: Characteristics and Caching Implications

Understanding the nature of the workloads and system demands created by users of the World Wide Web is crucial to properly designing and provisioning Web services. Previous measurements of Web client workloads have been shown to exhibit a number of characteristic features; however, it is not clear how those features may be changing with time. In this study we compare two measurements of Web client workloads separated in time by three years, both captured from the same computing facility at Boston University. The older dataset, obtained in 1995, is well-known in the research literature and has been the basis for a wide variety of studies. The newer dataset was captured in 1998 and is comparable in size to the older dataset. The new dataset has the drawback that the collection of users measured may no longer be representative of general Web users; however using it has the advantage that many comparisons can be drawn more clearly than would be possible using a new, different source of measurement. Our results fall into two categories. First we compare the statistical and distributional properties of Web requests across the two datasets. This serves to reinforce and deepen our understanding of the characteristic statistical properties of Web client requests. We find that the kinds of distributions that best describe document sizes have not changed between 1995 and 1998, although specific values of the distributional parameters are different. Second, we explore the question of how the observed differences in the properties of Web client requests, particularly the popularity and temporal locality properties, affect the potential for Web file caching in the network. We find that for the computing facility represented by our traces between 1995 and 1998, (1) the benefits of using size-based caching policies have diminished; and (2) the potential for caching requested files in the network has declined.

Load Balancing a Cluster of Web Servers using Distributed Packet Rewriting

In this paper, we propose and evaluate an implementation of a prototype scalable web server. The prototype consists of a load-balanced cluster of hosts that collectively accept and service TCP connections. The host IP addresses are advertised using the Round Robin DNS technique, allowing any host to receive requests from any client. Once a client attempts to establish a TCP connection with one of the hosts, a decision is made as to whether or not the connection should be redirected to a different host---namely, the host with the lowest number of established connections. We use the low-overhead Distributed Packet Rewriting (DPR) technique to redirect TCP connections. In our prototype, each host keeps information about connections in hash tables and linked lists. Every time a packet arrives, it is examined to see if it has to be redirected or not. Load information is maintained using periodic broadcasts amongst the cluster hosts.

Connection Scheduling in Web Servers

Under high loads, a Web server may be servicing many hundreds of connections concurrently. In traditional Web servers, the question of the order in which concurrent connections are serviced has been left to the operating system. In this paper we ask whether servers might provide better service by using non-traditional service ordering. In particular, for the case when a Web server is serving static files, we examine the costs and benefits of a policy that gives preferential service to short connections. We start by assessing the scheduling behavior of a commonly used server (Apache running on Linux) with respect to connection size and show that it does not appear to provide preferential service to short connections. We then examine the potential performance improvements of a policy that does favor short connections (shortest-connection-first). We show that mean response time can be improved by factors of four or five under shortest-connection-first, as compared to an (Apache-like) size-independent policy. Finally we assess the costs of shortest-connection-first scheduling in terms of unfairness (i.e., the degree to which long connections suffer). We show that under shortest-connection-first scheduling, long connections pay very little penalty. This surprising result can be understood as a consequence of heavy-tailed Web server workloads, in which most connections are small, but most server load is due to the few large connections. We support this explanation using analysis.

Measuring Web Performance in the Wide Area

One of the most vexing questions facing researchers interested in the World Wide Web is why users often experience long delays in document retrieval. The Internet's size, complexity, and continued growth make this a difficult question to answer. We describe the Wide Area Web Measurement project (WAWM) which uses an infrastructure distributed across the Internet to study Web performance. The infrastructure enables simultaneous measurements of Web client performance, network performance and Web server performance. The infrastructure uses a Web traffic generator to create representative workloads on servers, and both active and passive tools to measure performance characteristics. Initial results based on a prototype installation of the infrastructure are presented in this paper.

Popularity-Aware GreedyDual-Size Web Proxy Caching Algorithms

Web caching aims to reduce network traffic, server load, and user-perceived retrieval delays by replicating "popular" content on proxy caches that are strategically placed within the network. While key to effective cache utilization, popularity information (e.g. relative access frequencies of objects requested through a proxy) is seldom incorporated directly in cache replacement algorithms. Rather, other properties of the request stream (e.g. temporal locality and content size), which are easier to capture in an on-line fashion, are used to indirectly infer popularity information, and hence drive cache replacement policies. Recent studies suggest that the correlation between these secondary properties and popularity is weakening due in part to the prevalence of efficient client and proxy caches (which tend to mask these correlations). This trend points to the need for proxy cache replacement algorithms that directly capture and use popularity information. In this paper, we (1) present an on-line algorithm that effectively captures and maintains an accurate popularity profile of Web objects requested through a caching proxy, (2) propose a novel cache replacement policy that uses such information to generalize the well-known GreedyDual-Size algorithm, and (3) show the superiority of our proposed algorithm by comparing it to a host of recently-proposed and widely-used algorithms using extensive trace-driven simulations and a variety of performance metrics.

Temporal Locality in Web Request Streams: Sources, Characteristics, and Caching Implications

Temporal locality of reference in Web request streams emerges from two distinct phenomena: the popularity of Web objects and the {\em temporal correlation} of requests. Capturing these two elements of temporal locality is important because it enables cache replacement policies to adjust how they capitalize on temporal locality based on the relative prevalence of these phenomena. In this paper, we show that temporal locality metrics proposed in the literature are unable to delineate between these two sources of temporal locality. In particular, we show that the commonly-used distribution of reference interarrival times is predominantly determined by the power law governing the popularity of documents in a request stream. To capture (and more importantly quantify) both sources of temporal locality in a request stream, we propose a new and robust metric that enables accurate delineation between locality due to popularity and that due to temporal correlation. Using this metric, we characterize the locality of reference in a number of representative proxy cache traces. Our findings show that there are measurable differences between the degrees (and sources) of temporal locality across these traces, and that these differences are effectively captured using our proposed metric. We illustrate the significance of our findings by summarizing the performance of a novel Web cache replacement policy---called GreedyDual*---which exploits both long-term popularity and short-term temporal correlation in an adaptive fashion. Our trace-driven simulation experiments (which are detailed in an accompanying Technical Report) show the superior performance of GreedyDual* when compared to other Web cache replacement policies.