M. Arlitt, C.L. Williamson, Web server workload characterization, the search for invariants, Proceedings of ACM SIGMETRICS Conference, Philadelphia, PA, May, 1996.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....many HTTP requests to the site. For example, several images may have to be retrieved to display the page that contains the results of the execution of an E business function. Past studies of WWW workloads concentrated on information provider sites and found several characteristics common to them [5,7,11,20]. Some of these characteristics deal with file size distributions, file popularity distribution, self similarity in Web tra#c, reference locality, and user request patterns. A number of studies of di#erent Web sites found file sizes to exhibit heavytailed distributions and object popularity to be ....

....di#erent Web sites found file sizes to exhibit heavytailed distributions and object popularity to be Zipf like. Other studies of di#erent Web site environments demonstrated long range dependencies in the user request process, in other words, strong correlations in the user requests. In particular, [7] identified ten workload properties, called invariants,across six di#erent data sets, which included di#erent types of information provider Web sites. Some of the most relevant invariants are: i) images and HTML files account for 90 100 of the files transferred; ii) 10 of the documents account ....

[Article contains additional citation context not shown here]

M. Arlitt and C. Williamson, "Web Server Workload Characterization," Proc. 1996.

....Web sites found le sizes to have heavy tailed distributions and object popularity to be Zipf like. Other studies of di erent Web site environments demonstrated longrange dependencies in the user request process, primarily resulting from strong correlations in the user requests. In particular, [4] identi ed ten workload properties, called invariants, across six di erent data sets, which included different types of information provider Web sites. Some of the most relevant invariants are: i) images and HTML les account for 90 100 of the les transferred; ii) 10 of the documents account ....

....references on workload characterization in the WWW focus only on information provider Web sites [13] A few references have addressed the problem of workload characterization for e commerce sites. Existing work concentrates on characterizing Web workloads composed of sequences of le requests [4, 7]. The characteristics and statistical properties of workloads on the Web have been studied by many papers [3, 4, 7, 13] A number of studies of di erent sites identi ed WWW workload properties and invariants. For instance, le sizes have heavy tailed distributions (e.g. Pareto distribution) ....

[Article contains additional citation context not shown here]

M. Arlitt and C. Williamson, \Web Server Workload Characterization," Proc. 1996.

....Web sites found file sizes to have heavy tailed distributions and object popularity to be Zipf like. Other studies of different Web site environments demonstrated longrange dependencies in the user request process, primarily resulting from strong correlations in the user requests. In particular, [4] identified ten workload properties, called in variants, across six different data sets, which included different types of information provider Web sites. Some of the most relevant invariants are: i) images and HTML files account for 90 100 of the files transferred; ii) 10 of the documents ....

....references on workload characterization in the WWW focus only on information provider Web sites [13] A few references have addressed the problem of workload characterization for e commerce sites. Existing work concentrates on characterizing Web workloads composed of sequences of file requests [4, 7]. The characteristics and statistical properties of workloads on the Web have been studied by many papers [3, 4, 7, 13] A number of studies of differ ent sites identified WWW workload properties and invariants. For instance, file sizes have heavy tailed distributions (e.g. Pareto distribution) ....

[Article contains additional citation context not shown here]

M. Arlitt and C. Williamson, "Web Server Workload Characterization," Proc. 1996.

....indicate that the summary cache enhanced ICP solves the overhead problem of ICP, requires minimal changes, and enables scalable Web cache sharing over a wide area network. VIII. RELATED WORK Web caching is an active research area. There are many studies on Web client access characteristics [12] [4], 16] 36] 25] Web caching algorithms [50] 38] 10] as well as Web cache consistency [30] 34] 37] 15] Our study does not address caching algorithms or cache consistency maintenance, but overlaps some of client traffic studies in our investigation of the benefits of Web cache ....

M. Arlitt and C. Williamson, "Web server workload characterization," in Proc.

....the number of pages, their size and access probabilities. In our experiments, we set the number of client processes as 30 and used two different workloads, described in Table 1. The parameters that define workload WB are representative of the kinds of workload typically found in busy WWW servers [2]. 5 Results This section discusses the results obtained for both user and kernel level approaches. Recall that the performance metric is the average latency of a request as perceived by the server. Also, we used a fixed number of client processes (thirty) and for the kernel level approach, we ....

Arlitt, M. and Williamson, C., Web Server Workload Characterization, Proceedings of the 1996 SIGMETRICS Conference on Measurement and Modeling of Computer Systems, May 1996.

....the client machines used were SunSparc 20 with 256 Mbytes RAM running Solaris 2.5.1 and a Sun Enterprise 3000 with 512 MB running Solaris 2.5.1. The local network was a dedicated 10 Mbps Ethernet. The number of WWW clients were evenly divided among the machines. 4. Workload According to [Arlitt], the following aspects can characterize a Web server workload: arrival rate of requests; file size distribution at the server; average size of the requested documents; reference concentration; type of the objects (HTML, image, etc. and . percentage of successful requests. The ....

....following aspects can characterize a Web server workload: arrival rate of requests; file size distribution at the server; average size of the requested documents; reference concentration; type of the objects (HTML, image, etc. and . percentage of successful requests. The author [Arlitt] also enumerated ten workload characteristics that are common to Internet Web servers. For example, 90 of HTTP requests are for HTML or image files, which have a typical size of 10 Kbytes. Another invariant is that only 10 percent of the files respond to 90 of the HTTP requests and 90 of the ....

[Article contains additional citation context not shown here]

M. Arlitt and C. Williamson, "Web Server workload characterization", Proceedings of 1996-- SIGMETRICS Conference on Measurement and Modeling of Computer Systems, May 1996.

....Starting Date Duration Data Served cache.kaist.ac.kr 1 Apr 97 6 months 692.2G(bytes) Tab. 2. 1: Datasets of WWW object requests to the local cache Due to the easiness of incorporating multimedia data into a Web document, the high degree of variation in object type is observed in Web traffic [2] [8] 16] Because the Web cache server record the URL of the object being requested and the file extension and Multipurpose Internet Mail Extensions (MIME) table can be used categorizing document media type, it is possible to determine the relative request frequency for text, images, audio, video ....

M. Arlitt and C. Williamson. Web Server Workload Characterization. In Proceedings of ACM SIGMETRICS'96, San Francisco, USA, November 1996.

....conditions on the internet and the Web server by reducing the frequency of requests. The average file size that is transferred from a Web server is 8 10KB while the median file size is around 1.5 2. 0 KB according to a study done in [20] In another Web workload characterisation study done [2], file sizes in the range of 13 KB was the average reported for most of the sites. So a decision was made to have a document set of file sizes 1KB, 10KB, and 100KB from these sites. Average document transfer rates are around 5000 bytes per second on the internet [24] For the file sizes mentioned ....

Martin F. Arlitt and Carey L. Williamson, "Web Server Workload Characterization", Proceedings of the ACM SIGMETRICS Conference on Measurement and Modelling of Computer Systems, May 1996.

....the drawback of being interpreted by a Java Virtual Machine [19] 20] Web servers can easily be configured to deliver static pages to high number of concurrent users without substantial performance degradation. Moreover, performance analysis of Web servers for static requests has been studied [1] [6] 7] 13] and modeled [6] 12] However, Web server performance for dynamic documents has neither been experimentally measured nor thoroughly analyzed. In order to determine the technology that gives best performance, the costs and benefits of each need to be weighed. Benchmark tools for ....

....Analysis Studies such as [6] look at the evolving standards for performance measuring, monitoring, and benchmarking Web servers. A lot of work had been done on looking for the characteristics of the Web that makes it different from traditional distributed systems. Arlitt and Williamson in [1] [2] did a workload characterization for Web servers. Basically, their study showed the types of objects available at Web servers and analyzed relevant aspects such as file size distribution. Woodruf et al. [7] examined characteristics of a large number of HTML documents. Their results showed ....

Arlitt, M. and Williamson, C., Web Server workload characterization, Proceedings of the 1996 SIGMETRICS Conference on Measurement and Modeling of Computer Systems, 1996.

....study by submitting requests with different cookies to better understand if and how varying the cookie changes the response. 3. 3 Content Classification In analyzing the retrieved resources, we concentrate on results correlated primarily to resource content type using the classification given in [1] as a guide. In cases where server redirection has occurred, we classify a resource based on the resulting URL. We use the Content Type header as the first key in our categorization and, where appropriate, further classify the content based on other characteristics. The base categories we use are ....

....of hit counts. 6 Related Work There has been much related work on both web characterization and caching, but none that has focused specifically on characterization for improved caching. We have drawn on other web characterization studies in trying to understand and classify our results [1, 7, 8, 15, 22]. Previous work we and others have done on web caching [3, 5, 13] has motivated this work in trying to better understand the potential of web caching. Kroeger, et al. [14] published a previous study on the potential of caching and prefetching in reducing web latency. This study was based on a proxy ....

M. Arlitt and C. Williamson. Web server workload characterization. In Proceedings of the ACM SIGMETRICS International Conference on Measurement and Modeling of Computer Systems, May 1996.

....of user interests. 6 Related Work There has been much related work on both web characterization and caching, but none that has focused specifically on characterization for improved caching. We have drawn on other web characterization studies in trying to understand and classify our results [1, 10, 12, 15, 18]. Previous work we and others have done on web caching [4, 6, 13] has motivated this work in trying to better understand the potential of web caching. Kroeger, et al. [14] published a previous study on the potential of caching and prefetching in reducing web latency. That study was based solely on ....

M. Arlitt and C. Williamson. Web server workload characterization. In Proceedings of the ACM SIGMETRICS International Conference on Measurement and Modeling of Computer Systems, May 1996.

....type of workload. 1.0 Introduction In the past decade, the popularity of the World Wide Web has grown exponentially. By the year 2000, the number of hosts on the web is expected to pass a hundred million [1] This explosive growth has to led to many studies of web site content and access patterns [2,3,4,5]. However, most of these studies have focused on HTTP logs from more traditional web sites, sites with small amounts of storage and relatively small files. At the same time, institutions with large archives of documents (for example, museums and libraries) are beginning to digitize their holdings ....

....images or least accessed parts of images. The trade off is the user s download time, which must be kept low using caching. Exploring these issues further is left for future work. 6.0 Related Work There are many studies on access patterns of web sites. We reference only a few of them here [2,3,4,5]. Our study differs from these in several ways; the size of the objects stored on the site, and the tile based nature of the content. To our knowledge, we present the first study of access patterns for a web service that has both a large number of large images and a tile based approach to ....

Arlitt, F. Williamson, C.L. Web Server Workload Characterization, The Search for Invariants. Proceedings of the 1996 SIGMETRICS Conference on Measurement and Modeling of Computer Systems, Philadelphia, PA, USA, 23-26 May 1996.

....of user interests. 6 Related Work There has been much related work on both web characterization and caching, but none that has focused specifically on characterization for improved caching. We have drawn on other web characterization studies in trying to understand and classify our results [1, 9, 11, 14, 17]. Previous work we and others have done on web caching [3, 5, 12] has motivated this work in trying to better understand the potential of web caching. Kroeger, et al. [13] published a previous study on the potential of caching and prefetching in reducing web latency. That study was based solely on ....

M. Arlitt and C. Williamson. Web server workload characterization. In Proceedings of the ACM SIGMETRICS International Conference on Measurement and Modeling of Computer Systems, May 1996.

....the number of pages, their size and access probabilities. In our experiments, we set the number of client processes as 30 and use two different workloads, described in Table 1. The parameters that define workload WB are representative of the kinds of workload typically found in busy WWW servers [2]. 5 Results This section discusses the results obtained for both user and kernel level approaches. Recall that the performance metric is the average latency of a request as perceived by the server. We use 30 client processes and have 15 processes issue requests of type A (higher priority) and the ....

Arlitt, M. and Williamson, C., Web Server Workload Characterization, Proceedings of the 1996 SIGMETRICS Conference on Measurement and Modeling of Computer Systems, May 1996.

....results indicate that the summary cache enhanced ICP solves the overhead problem of ICP, requires minimal changes, and enables scalable Web cache sharing over a wide area network. 8 Related Work Web caching is an active research area. There are many studies on Web client access characteristics [10, 3, 14, 33, 23], web caching algorithms [49, 35, 8] as well as Web cache consistency [28, 31, 34, 13] Our study does not address caching algorithms or cache consistency maintanence, but overlaps some of client traffic studies in our investigation of the benefits of Web cache sharing. Recently, there have been a ....

M. Arlitt and C. Williamson. Web server workload characterization. In Proceedings of the 1996 ACM SIGMETRICS International Conference on Measurement and Modelling of Computer Systems, May 1996.

....of user interests. 6 Related Work There has been much related work on both web characterization and caching, but none that has focused specifically on characterization for improved caching. We have drawn on other web characterization studies in trying to understand and classify our results [1, 8]. Previous work we and others have done on web caching [3, 5, 12] has motivated this work in trying to better understand the potential of web caching. Kroeger, et al. [14] published a previous study on the potential of caching and prefetching in reducing web latency. This study was based on a ....

M. Arlitt and C. Williamson. Web server workload characterization. In Proceedings of the ACM SIGMETRICS International Conference on Measurement and Modeling of Computer Systems, May 1996. ftp://ftp.cs.usask.ca/pub/discus/paper.96-3.ps.Z.

No context found.

M. Arlitt, C.L. Williamson, Web server workload characterization, the search for invariants, Proceedings of ACM SIGMETRICS Conference, Philadelphia, PA, May, 1996.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC