M. Arlitt. A Performance Study of Internet Web Servers. Master's Thesis, Uni. of Saskatchewan, Canada, 96  Home/Search   Document Details and Download   Summary   Related Articles   Check

....consumption on future equivalent requests. In this report, the focus will be on the document replacement policies used by the web cache manager to control the content of its cache. According to several studies, document replacement strategies have great impact on the web cache performance [18] [3]. This process is tightly related to the cache size, it is obvious if an infinite cache exists, there is no need to such a document removal process all the documents that enter the cache are keep for ever. Contrarily, in real situations documents have to be replaced because of a lack of memory ....

Martin F. Arlitt, "A Performance study of Internet Web Servers", Master Thesis, Departement of computer Science, University of Saskatchewan.

....issue. There are many cache replacement policies proposed in the literature. An algorithm called Least Recently Used (LRU) 1, 4, 5] uses recency of use information for each item in the cache to remove the least recently used item, while another popular algorithm called Least Frequently Used (LFU) [2, 3, 4, 5, 12, 13] 169 Mudashiru Busari maintains a count of the number of references to each item in the cache, and removes the document with the lowest count when a replacement decision needs to be made. If two documents have the same reference count it then uses the LRU policy to break the tie. Frequency based ....

....with the lowest count when a replacement decision needs to be made. If two documents have the same reference count it then uses the LRU policy to break the tie. Frequency based algorithms, such as LFU, have been found to outperform recency based algorithms such as LRU, on Web server workloads [2, 3, 4]. LFU has many variations; for example, LFU Aging, LFU and LFU Aging were studied in [3] The workload characterization study of Web servers in [4] shows that about 15 40 of resources on Web servers are accessed only once, which means caching such documents is not worthwhile. The LFU algorithm ....

M. Arlitt, "A Performance Study of Internet Web Servers," M.Sc. Thesis, Department of Computer Science, University of Saskatchewan, 1996.

....the signal, it is called quasiperiodic. The analysis done here shows that the data is quasiperiodic; however, for simplicity we always use the term periodic instead of quasiperiodic. 4.2. 3 Web tra#c characterization and self similarity The main objective of the study reported by Arlitt [Arl96] and Arlitt and Williamson [AW96b] was to identify workload invariants across a set of log files collected from di#erent locations. The authors identified 10 invariants in all collected workloads. Self similarity was not reported as an invariant across all workloads, however some of the tested ....

M. F. Arlitt. A performance study of internet web servers. Master's thesis, University of Saskatchewan, Saskatoon, Saskatchewan, May 1996.

....employed at di erent locations on the Web. At the client side [2, 18, 23, 33, 62, 74] browser software can utilize memory (non persistent cache) or disk cache (persistent cache) to store copies of recently requested documents to satisfy subsequent access to the same document. At the server side [6, 7, 8, 9, 18], recently requested documents can be stored in memory to satisfy future requests to the same document by the same user or another user, thus eliminating disk I O access time. Within the network [1, 10, 12, 13, 18, 20, 28, 56, 59, 62, 65, 74, 76] proxy caches can be employed to act as ....

....addressed in a caching proxy in order to improve Web performance. A Web proxy needs to make good replacement decisions each time a new document is coming into the cache. After deciding that a new document is to be cached, some of the old ones 1 These documents are referred to as one timers [8]. 3 in the cache might need to be evicted to create enough storage space for the new document. It is therefore important to make good choices about which document (or documents) to remove (e.g. to ensure that only documents that are not likely to be requested in the near future are evicted from ....

[Article contains additional citation context not shown here]

M. Arlitt, A Performance Study of Internet Web Servers, M.Sc. Thesis, Department of Computer Science, University of Saskatchewan, 1996.

....how workload characteristics have changed over time. The statistical analysis of the new data set shows that most of the Web characteristics outlined in their previous study can still be modelled using power law distributions. Many researchers have studied Web server workload characteristics [2, 4, 5, 16, 54]. In an extensive study, Arlitt and Williamson [5] identi ed ten common workload characteristics using six di erent Web server traces that spanned durations from one week to one year. The characteristics identi ed are: a) approximately 90 of all requests result in successful transfer of ....

....for at least 70 of the references to the servers; and (j) Web servers are accessed by thousands of domains, with 10 of the domains accounting for more than 75 of the usage. Abdulla et al. 1] identi ed nine common workload characteristics for Web proxies, similar to the Web server study in [4, 5]. The characteristics identi ed are: a) the mean le size is approximately 2 KB; b) the median le size is less than 27 KB; c) HTML, image and dynamic les account for 90 98 of the les accessed; d) image les account for the most bytes transferred; e) less than 11 of the accesses are to ....

[Article contains additional citation context not shown here]

M. Arlitt, A Performance Study of Internet Web Servers, M.Sc. Thesis, Department of Computer Science, University of Saskatchewan, 1996. Available at URL: http://www.cs.usask.ca/ftp/pub/ discus/thesis arlitt bw.ps.Z

....cached documents in the same way as LFU. The difference is that LFU Aging additionally monitors the average of the reference counts of all cached documents. When the average is given over a given maximum number, the cache starts again. In our simulations, the maximum is set to 10 as suggested in [4]. Size: The SIZE algorithm sorts cached documents by size. Documents with the same size are sorted by recency. When there is insufficient space for caching the most recently requested document, the least recently used of the document with the largest size is replaced. More than one document will ....

M Arlitt, A performance study of Internet Web servers, M.Sc. Thesis, University of Saskatchewan , 1996.

....or more requests went to only 84 servers (13 of the top 20 servers were outside the vt.edu domain and would benefit from caching) Fig. 1 suggests that the number of requests to each server in workload BL follows a Zipf distribution. In comparison, the requested URL in traces studied elsewhere [4, 9] also follows a Zipf distribution. 1 10 100 1000 10000 100000 1 10 100 1000 10000 Server: Ranked by number of requests Backbone Local Traffic for Sep17 Oct31 1995 Figure 1: Distribution of requests for particular servers for workload BL. Many clients in Fig. 1 are requesting URLs from ....

M. F. Arlitt. A performance study of internet web servers. Master's thesis, Computer Sci. Dept., University of Saskatchewan, Saskatoon, Saskatchewan, May 1996.

....to represent the signal, it is called quasiperiodic. The analysis done here show that the data is quasiperiodic; however, for simplicity we always use the term periodic instead of quasiperiodic. 2. 3 Web tra#c characterization and self similarity The main objective of the study reported by Arlitt [12] and Arlitt and Williamson [13] was to identify workload invariants across a set of log files collected from di#erent locations. The authors identified 10 invariants in all collected workloads. Self similarity was not reported as an invariant across all workloads, however some of the tested ....

M. F. Arlitt. A performance study of Internet Web servers. Master's thesis, University of Saskatchewan, Saskatoon, Saskatchewan, May 1996.

....or more requests went to only 84 servers (13 of the top 20 servers were outside the vt.edu domain and would benefit from caching) Fig. 1 suggests that the number of requests to each server in workload BL follows a Zipf distribution. In comparison, the requested URL in traces studied elsewhere [4, 9] also follows a Zipf distribution. 1 10 100 1000 10000 100000 1 10 100 1000 10000 Server: Ranked by number of requests Backbone Local Traffic for Sep17 Oct31 1995 Figure 1: Distribution of requests for particular servers for workload BL. Many clients in Fig. 1 are requesting URLs from ....

M. F. Arlitt. A performance study of internet web servers. Master's thesis, Computer Sci. Dept., University of Saskatchewan, Saskatoon, Saskatchewan, May 1996.

....We present results for a workload generated by WebStone [21] which is a configurable tool for benchmarking Web server performance, available from Silicon Graphics. We parameterized the server workload generated by WebStone to capture the heterogeneous nature of HTTP requests, using values from [4]. Specifically, we used a file size distribution with a heavy tail to capture the fact that Web servers must concurrently handle some requests for huge multimedia files and a large number of requests for small HTML and image documents. Such distributions occur in the size of files requested at ....

....probabilities. A request for a page means a request for each file that makes up the page. Table 1 gives baseline information for the HTTP workload used in our experiments. The parameters that define the workload are representative of the kinds of workload typically found in busy WWW servers [4]. File size (KBytes) Access probability Item Number of files Total Average Total Average HTML 24 180 7.5 0.192 0.008 Images 29 385 13.28 0.754 0.026 Sound 20 3580 179 0.05 0.0025 Video 4 9216 2304 0.004 0.001 Table 1: Characteristics of an HTTP Workload 4 Architecture of the Monitor Figure 1 ....

M. F. Arlitt. A performance study of Internet web servers. M.Sc. Thesis, Department of Computer Science, University of Saskatchewan, Saskatoon, Saskatchewan, June 1996.

....space at the proxy servers, it is often necessary to remove previously cached documents before storing any newly requested documents. In this paper, the focus will be on the replacement policies used in web caching. Document replacement strategies have a great impact on web cache performance [13, 2]. The replacement process is tightly related to the cache size: if an infinite cache exists, there is no need for such a document removal process since all the documents that enter the cache are kept for ever. In real caches documents have to be replaced because of a lack of memory space, and the ....

Arlitt, M. F., "A Performance study of Internet Web Servers," Master Thesis, Departement of computer Science, University of Saskatchewan.

....or more requests went to only 84 servers (13 of the top 20 servers were outside the vt.edu domain and would benefit from caching) Fig. 1 suggests that the number of requests to each server in workload BL follows a Zipf distribution. In comparison, the requested URL in traces studied elsewhere [4, 9] also follows a Zipf distribution. 1 10 100 1000 10000 100000 1 10 100 1000 10000 Server: Ranked by number of requests Backbone Local Traffic for Sep17 Oct31 1995 Figure 1: Distribution of requests for particular servers for workload BL. Many clients in Fig. 1 are requesting URLs from a few ....

M. F. Arlitt. A performance study of internet web servers. Master's thesis, Computer Sci. Dept., University of Saskatchewan, Saskatoon, Saskatchewan, May 1996.

.... Recently Used (LRU) Least Frequently Used (LFU) Least Frequently Used with Aging (LFU Aging) and Frequency Based Replacement (FBR) In addition, two new approaches are considered: size based replacement, as proposed by Williams et al. 20] and a modified version of LFU, proposed by Arlitt [3]. The LRU policy removes from the cache the file that has not been used for the longest period of time. LRU works well in reference streams that exhibit the property of temporal locality (this property suggests that files that have been referenced in the recent past will likely be referenced again ....

....LFU policy was employed, so that the reference counts, over time, can be reduced back to one, making more documents candidates for replacement. This second new policy is denoted as LFU Aging. For the purpose of comparison, both LFU Aging and LFU Aging are tested with the same aging parameters [3]. 3.1.3 Threshold Policy The third factor used in this study is a threshold policy that places constraints on the sizes of documents allowed in the cache. A requested document is added to the file cache if and only if the size of the requested document is within the acceptable range. One ....

[Article contains additional citation context not shown here]

M. Arlitt, "A Performance Study of Internet Web Servers", M.Sc. Thesis, Department of Computer Science, University of Saskatchewan, June 1996.

No context found.

M. Arlitt. A Performance Study of Internet Web Servers. Master's Thesis, Uni. of Saskatchewan, Canada, 96

No context found.

# M.F. Arlitt, "A Performance Study of Internet Web Servers," master 's thesis, Computer Science Dept., Univ. of Saskatchewan, Saskatoon, Saskatchewan, Canada, 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