A. K. Iyengar, M. S. Squillante, and L. Zhang. Analysis and characterization of large-scale Web server access patterns and performance. World Wide Web, 2(1-2):85--100, June 1999.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....to several decades ago [JMF99] most of the data sets were treated as independent samples, while the temporal correlation between neighboring samples are largely unexploded. Classical time series analysis techniques has been widely used in many domains such as financial data and web stat analysis [ISZ99] where the problem of identifying seasonality reduces to the problem of parameter estimation in known order of ARMA model, and the order is determined with statistical tests on the correlation functions, yet this model does not readily adapt to domains with frequent discontinuities. New ....

Arun Iyengar, Mark S. Squillante, and Li Zhang. Analysis and characterization of large-scale web server access patterns and performance. World Wide Web, 2(1-2):85-- 100, 1999.

....a load that is approximately 58 of the processor s capacity at its highest frequency setting. Policies based on dynamic voltage scaling are not very effective at low workload intensities. However, previous studies have observed that Web servers are relatively idle for large fractions of time [13]. Even when idle, server processors consume significant amounts of power. Unfortunately, since incoming requests arrive asynchronously, web servers cannot afford to use energy conserving states with significant wakeup penalties such as the hibernation mode commonly found in laptops. Request ....

A. Iyengar, M. Squillante, and L. Zhang. Analysis and characterization of large-scale web server access patterns and performance. World Wide Web, 2(12) :85-100, June 1999.

....demand shows significant, and predictable, variations in the hourly scales. These results will provide guidance for the design of adaptive policies. I. INTRODUCTION The infrastructure supporting many Internet services and the clients of these services are globally distributed (e.g. Olympics96 [1], WorldCup98 [2] CDNs [3] 4] Since the client population is practically unbounded for many services, demand for these services can vary over a wide range with peak demand one or two orders of magnitude larger than average demand. It is not cost effective to statically provision the service ....

....Data Center [6] Work done during internship at Hewlett Packard Laboratories To facilitate the design of an effective adaptive distributed service infrastructure, it is essential to develop a good understanding of the dynamic variations in the service demand. While several previous studies [1], 7] 2] 8] 9] characterize demand for Internet services, they are primarily useful in designing a single site for the service and in caching studies. They do not examine workload properties that are crucial in adaptively allocating distributed resources. Other studies [10] 11] cluster ....

[Article contains additional citation context not shown here]

A. Iyengar, M. S. Squillante, and L. Zhang, "Analysis and characterization of large-scale web server access patterns and performance," World Wide Web, vol. 2, no. 1-2, pp. 85--100, 1999.

....APPROACH 0 10 20 30 40 50 60 70 80 90 100 0 5 10 15 20 25 Time of day Fig. 1. Scatter plot of daily access rate (number of web requests per minute) over 60 days of a web trace from the Internet Traffic Archives. Client Workload Characterization: Web server traffic has been studied in [11], where they model the number of hits at a web server in each small time interval (5 minutes) as a stationary ARIMA time series. Other stationary stochastic models are found in [17] and [7] In the same spirit we model the client resource requirement as a stationary stochastic process. Figure 1 ....

Arun K. Iyengar, Mark S. Squillante, and Li Zhang, Analysis and Characterization of Large-Scale Web Server Access Patterns and Performance, The 8th World Wide Web Conference, 1999.

....than old ones. 4 Characterizing the request flow As the test objects of this study, we have chosen several activity logs for various web servers. We tried to diversify the phenotypes of the traces to probe for the limits of applicability of our algorithms. It is well known (see e.g. JKR02, ISZ99] that a load based taxonomy of web server traffic is both an extremely important and an extremely challenging problem. The behavior of the traffic is shaped by a combination of so many technological, sociological, psychological (to name a few) factors that a quantification of even basic patterns ....

....varied along the x axis and the y axis plots the probability that the number of requests in a slot exceeds this value. Note the heavy tail properties of the Olympics and World Cup distributions, which are given in log log plots (the plot for the Olympics data confirms the log normal estimate of [ISZ99] By contrast, the distribution for the NASA data is relatively flat. As is common in prediction problems, the performance of the prediction algorithms strongly depends on the power spectrum (see [GS92] for example) that is, the absolute value of the Fourier transform of the time series. ....

A. Iyengar, M. Squillante, and L. Zhang. Analysis and Characterization of Large-Scale Web Server Acess Patterns and Performance. World Wide Web, 2, June 1999.

....trac is very high. The World Cup Web site study [5] presents a peak workload analysis pointing out that le referencing is more concentrated on a few extremely popular pages. Also many clients repeatedly visited the site with shorter inter session time. A study of the 1998 Olympic Games Web site [16] analyzed the workload and developed trac models that could be used to predict seasonal trac variations and peak request rates at a Web site during its normal operation. Neither study is however, concerned with DoS attacks or implications on CDNs. A recent study [8] examined IP ow level trac at ....

A. K. Iyengar, M. S. Squillante, and L. Zhang. Analysis and characterization of large-scale Web server access patterns and performance. World Wide Web, June 1999.

....is very high. The World Cup Web site study [5] presents a peak workload analysis pointing out that file referencing is more concentrated on a few extremely popular pages. Also many clients repeatedly visited the site with shorter inter session time. A study of the 1998 Olympic Games Web site [16] analyzed the workload and developed traffic models that could be used to predict seasonal traffic variations and peak request rates at a Web site during its normal operation. Neither study is however, concerned with DoS attacks or implications on CDNs. A recent study [8] examined IP flow level ....

A. K. Iyengar, M. S. Squillante, and L. Zhang. Analysis and characterization of large-scale Web server access patterns and performance. World Wide Web, June 1999.

....known for the next L time intervals (following the current interval) In reality, lookahead will require that some forecasting mechanism be used to predict future demands. Fortunately, web trac exhibits properties of predictability over the short term which render this practical as demonstrated in [13, 18]. We measure the performance of an online allocation algorithm using its competitive ratio; that is, the maximum over all instances of the ratio of the bene t gained by the optimal o ine allocation to the bene t gained by the online allocation on the same instance. Since we are dealing with a ....

A.K. Iyengar, Mark S. Squillante and L. Zhang, \Analysis and Characterization of Large-Scale Web Server Access Patterns and Performance". World Wide Web, 2(1999), pp. 85-100.

....processed per unit time. We analyzed a server trace obtained from the server web site for Olympics 1998. This workload was generated by a major Sporting and Event web site hosted 25 by IBM, which in 1998 server 56.8 million requests on the peak day, 12 of which were to dynamically generated data [37]. The peak number of requests serviced by a server in unit time, can be approximated to the peak capacity of that server during that interval. The spare capacity available for prefetching in each time interval can be viewed as the di erence of this maximum and the actual requests processed during ....

A. Iyengar, M. Squillante, and L. Zhang. Analysis and characterization of large-scale web server access patterns and performance. In World Wide Web, June, 1999.

....the web server tends to be consistent in a short time window. For example, the coef cientofvariation (CoV) of access rates decrease to around 1 when they are measured on an hourly basis. The decrease for CoV is even more obvious during busy hours. Simi546 lar observations have been reported in [24, 19, 23], which suggest that a multi state Modulated MarkovPoisson Process (MMPP) can be used to approximate or predict the burstiness of the aggregate input of the web server, whichis discussed in more detail in Section 3.3. Note that during a week s observation period, there is one overload point ....

....priority group is based on the prediction of the request rate of incoming prioritized tasks. Apparent seasonal workload patterns corresponding to daily cycles discussed in the previous section can be used to predict currenttracintensity based on the workload history. On the other hand, reports in [19, 23] suggested that the aggregate web trac tends to smooth out as Poisson trac in short observation time windows. This fact was further proved by Morris and Lin in [24] Based on the above published results, we decided to use MarkovModulated Poisson Process (MMPP) described in [16] to capture the ....

A. K. Iyengar, M. S. Squillante, and L. Zhang. Analysis and characterization of large-scale web server access patterns and performance. World Wide Web, pages 85-100, 1999.

....the web server tends to be consistent in a short time window. For example, the coef cient of variation (CoV) of access rates decrease to around 1 when they are measured on an hourly basis. The decrease for CoV is even more obvious during busy hours. Simi lar observations have been reported in [24, 19, 23], which suggest that a multi state Modulated Markov Poisson Process (MMPP) can be used to approximate or predict the burstiness of the aggregate input of the web server, which is discussed in more detail in Section 3.3. Note that during a week s observation period, there is one overload point ....

....group is based on the prediction of the request rate of incoming prioritized tasks. Apparent seasonal workload patterns corresponding to daily cycles discussed in the previous section can be used to predict current trac intensity based on the workload history. On the other hand, reports in [19, 23] suggested that the aggregate web trac tends to smooth out as Poisson trac in short observation time windows. This fact was further proved by Morris and Lin in [24] Based on the above published results, we decided to use MarkovModulated Poisson Process (MMPP) described in [16] to capture the ....

A. K. Iyengar, M. S. Squillante, and L. Zhang. Analysis and characterization of large-scale web server access patterns and performance. World Wide Web, pages 85-100, 1999.

....But IPB does not include enough traffic types and updated models of existing applications. Almeida 1998] provided a model, called the Wisconsin Proxy Benchmark (WPB) This model can simulate the request streams according to the temporal locality patterns that are common to Web proxy servers. [Iyengar 1998] developed a tool called Flintstone. This model uses statistical methods to isolate and characterize the trends, interdependencies, seasonal behavior and noise in the access patterns. Flintstone provides an effective approach for predicting peak request rates for analyzing and characterizing Web ....

Arun K. Iyengar, et al, Analysis and Characterization of Large Scale Web Server Access Patterns and Performance, (also in World Wide Web, June 1999), http://www.research.ibm.com/people/i/iyengar/arun2.html, 1998.

No context found.

A. Iyengar, M. Squillante, and L. Zhang. Analysis and characterization of large-scale web server access patterns and performance. In World Wide Web, June 1999.

No context found.

A. Iyengar, M. Squillante, and L. Zhang. Analysis and characterization of large-scale web server access patterns and performance. In World Wide Web, June 1999.

No context found.

Arun K. Iyengar, Mark S. Squillante and Li Zhang (1999). "Analysis and Characterization of Large-Scale Web Server Access Patterns and Performance." World Wide Web. http://citeseer.nj.nec.com/iyengar99analysis.html

....levels. This performance requirement is further compounded by bursty access patterns which are common to many popular Web sites, resulting in high peak to average request rates 1389 1286 02 see front matter 2002 Elsevier Science B.V. All rights reserved. PII:S1389 1286(01)00241 9 76 [16]. In order to handle high request rates, it is often necessary to use multiple processors. One technique for reducing the amount of hardware needed at a Web site and improving throughput is to place one or more high performance caches in front of the Web servers known as Web server accelerators or ....

A. Iyengar, M. Squillante, L. Zhang, Analysis and characterization of large-scale Web server access patterns and performance, World Wide Web 2 (1/2) (1999) 85 100.

....consistency for web sites serving large amounts of dynamically generated data. Our study is based on the workload generated by a major Sporting and Eventweb site hosted by IBM # , which in 1998 served 56.8 million requests on the peak day, 12 of whichwere to dynamically generated data [12]. The rst issue we address is scalability. In server driven consistency, scalability can be limited byanumber of factors: # As the number of clients increases, the amount of memory needed to keep track of the content of clients caches may become large. # Servers may experience bursts of load ....

A. Iyengar, M. Squillante, and L. Zhang. Analysis and characterization of large-scale web server access patterns and performance. In World Wide Web,June 1999.

....consistency for web sites serving large amounts of dynamically generated data. Our study is based on the workload generated by a major Sporting and Event web site hosted by IBM 1 , which in 1998 served 56.8 million requests on the peak day, 12 of which were to dynamically generated data [12]. The rst issue we address is scalability. In server driven consistency, scalability can be limited by a number of factors: As the number of clients increases, the amount of memory needed to keep track of the content of clients caches may become large. 1 The 1998 Olympic Games web site ....

A. Iyengar, M. Squillante, and L. Zhang. Analysis and characterization of large-scale web server access patterns and performance. In World Wide Web, June 1999.

....of Objects (logarithmic scale) Fig. 10. The distribution of the number of incoming edges for nodes of ODG s at the 2000 Olympic Games Web site. and the Windows NT (version 4.0) operating system. The distribution of Web pages sizes is similar to the one for the 1998 Olympic Games Web site [8] as well as more recent Web sites deploying our system; the average Web page size is around 10 Kbytes. Fragment sizes are typically several hundred bytes but usually less than 1 Kbyte. The distribution of fragment sizes is also representative of real Web sites deploying our system. Figure 13 ....

Arun Iyengar, Mark Squillante, and Li Zhang. Analysis and Characterization of Large-scale Web Server Access Patterns and Performance. In World Wide Web, June 1999.

No context found.

Iyengar, A. K., M. S. Squillante, and L. Zhang (1998b), "Analysis and Characterization of Large-Scale Web Server Access Patterns and Performance," Technical Report RC 21328, IBM Research Division.

No context found.

A. K. Iyengar, M. S. Squillante, and L. Zhang. Analysis and characterization of large-scale Web server access patterns and performance. World Wide Web, 2(1-2):85--100, June 1999.

No context found.

Arun Iyengar, Mark S. Squillante, and Li Zhang. Analysis and characterization of large-scale web server access patterns and performance. The World Wide Web Journal, 2:85--100, 1999.

No context found.

A. K. Iyengar, M. S. Squillante, and L. Zhang. Analysis and characterization of large-scale Web server access patterns and performance. World Wide Web, 2(1-2):85--100, June 1999.

No context found.

A. K. Iyengar, M. S. Squillante, and L. Zhang, "Analysis and characterization of large-scale web server access patterns and performance," World Wide Web, pp. 85-100, 1999.

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