G. Banga and P. Druschel. Measuring the capacity of a web server under realistic loads. World Wide Web, 2(1-2):69--83, 1999.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... resources (such as TCP control blocks or file descriptors) for all accepted requests, the client machines must additionally allocate resources for all connections which the server has repeatedly refused (due to full SYN queue) Existing web workload generators include Surge [9] and Sclient [6]. Surge mimics a closed system with several users, where each user makes a request and after receiving the response waits for a certain think time before it makes the next request. Note that in a closed system it is not possible to create overload, since a new request will be made only if another ....

G. Banga and P. Druschel. Measuring the capacity of a web server under realistic loads. World Wide Web, 2(1-2):69--83, 1999.

....server by requesting objects as quickly as possible. However, some benchmarks attempt to recreate more realistic arrival distributions. The SURGE benchmark uses the concept of user equivalents to recreate the active and idle periods of typical users. Another benchmark, Scalable Client (s client) [4], uses an open loop architecture to produce a fixed request arrival rate regardless of server load. Both these approaches cause the server to deal with the bursty arrival of requests, which can significantly affect server performance. Protocol version. Some benchmarks, such as SPECWeb99, have ....

....on the last TCP segment of each conversation, improving performance for small transfers. Requests Clients Responses Server Figure 1: Workload Generation 3.3 WWW Client Workload Generator Software We use two client HTTP workload generators to evaluate WWW server performance. We use s client [4] as a micro benchmark to generate large numbers of requests for the same single file on the server, which lets us see how performance changes as a function of the requested file size. We also use our own workload generator Waspclient as a macro benchmark to measure more realistic aggregate ....

[Article contains additional citation context not shown here]

Gaurav Banga and Peter Druschel. Measuring the capacity of a Web server under realistic loads. World Wide Web Journal (Special Issue on World Wide Web Characterization and Performance Evaluation), 2(1):69--83, May 1999.

.... for the success of many 84 proposed techniques (e.g. PPR96, Spe97] and implemented Web systems, including search and meta search engines (e.g. Inf02b, SE95, SE97, DH97c, HD97, Inf02a, ZE98, ZE99, LG98b, LG98a, Goo02, Alt02, Ink02, Fas02, McB94, Lyc02b] focused crawlers (e.g. CGMP98, CvdBD99, BSHJ 99, Men97, MB00, Lie97, RM99] linkage analyzers (e.g. BFJ96, DH99, KRRT99, FLGC02, CDR 99, BH98b, BP98, DGK IBM00, Kle98] and intelligent Web agents (e.g. AFJM95, JFM97, Mla96, Lie95, Lie97, MB00, BS97, LaM96, LaM97, Lie97, PP97, Dav99a] as we will describe below. While ....

....and description) and the quality of these page descriptors is thus quite important to an after the fact textual ranking or clustering of the pages. 5.3. 4 Focused crawlers Focused crawlers are Web crawlers that follow links that are expected to be relevant to the client s interest (e.g. CvdBD99, BSHJ 99, Men97, MB00, Lie97, RM99] and the query similarity crawler in [CGMP98] They may use the results of a search engine as a starting point, or they may crawl the Web from their own dataset. In either case, they assume that it is possible to find highly relevant pages using local search ....

[Article contains additional citation context not shown here]

Gaurav Banga and Peter Druschel. Measuring the capacity of a Web server under realistic loads. World Wide Web Journal, 2(1-2):69--83, 1999. Special Issue on World Wide Web Characterization and Performance Evaluation.

....Pareto distribution (with parameter ) The largest of the total load, exhibiting a strong heavytailed property. of files have size less than 1K bytes. of files have size less than 9.3K bytes. 4.1. 4 Generating requests at client machines In our experiments, we use sclient [6] for creating connections at the client machines. The original version of sclient makes requests for a certain file in periodic intervals. We modify sclient to read in traces and make the requests according to the arrival times and file names given in the trace. To create a particular load, we ....

Gaurav Banga and Peter Druschel. Measuring the capacity of a web server under realistic loads. World Wide Web, 2(12) :69--83, 1999.

....response latencies, but also discover unexpected caching bugs in two of the proxies tested. 1 Introduction There are many possible methods to evaluate caching algorithms and implemented proxy caches [16] Of those that examine real systems, the typical approach is to apply benchmark tests (e.g. [8, 5, 4, 9, 29, 23, 31]) to either measure performance with a standard load, or to determine the maximum load attainable. The current de facto process is to compare many proxies using the same Polygraph [31] workload (e.g. in a caching competition [32] Unfortunately, these approaches are unusable for evaluating proxy ....

G. Banga and P. Druschel. Measuring the capacity of a Web server under realistic loads. World Wide Web Journal, 2(1-2):69--83, 1999. Special Issue on World Wide Web Characterization and Performance Evaluation.

....characterizes the Web (the number of online users is growing at about 90 per annum) Many efforts have also been directed at improving the performance of a Web server node at the software level, namely software scale up. This includes the operating system and also the Web server application [14, 12, 57, 75, 80]. The Flash Web server ensures that its threads and processes are never blocked by using an asymmetric multiprocess event driven architecture [80] Nahum et al. 75] have analyzed how a general purpose operating system and the network protocol stack can be improved to provide support for ....

G. Banga and P. Druschel. Measuring the capacity of a Web server under realistic loads. World Wide Web, 2(1):69-83, May 1999.

....ff parameter 1:2) The largest 3 of the requests make up 50 of the total load, exhibiting a strong heavytailed property. 50 of files have size less than 1K bytes. 90 of files have size less than 9.3K bytes. 4.1. 4 Generating requests at client machines In our experiments, we use sclient [6] for creating connections at the client machines. The original version of sclient makes requests for a certain file in periodic intervals. We modify sclient to read in traces and make the requests according to the arrival times and file names given in the trace. To create a particular load, we ....

Gaurav Banga and Peter Druschel. Measuring the capacity of a web server under realistic loads. World Wide Web, 2(12) :69--83, 1999.

....extended with resource containers and with the resource management framework presented in this chapter. All experiments were performed on a 500MHz Pentium III machine con gured with 1 GB of main memory. The Web requests were generated by a HTTP client program designed for Web server benchmarking [16]. The program can generate HTTP requests from synthetic or real logs either as fast as the Web server can handle them or at a rate dictated by timestamps in the log. Seven 166 MHz Pentium Pro machines were used to run the 55 client program. The server machine and the seven client machines were ....

....to be directly sent to the clients. The second con guration shown in Figure 4.3 emulates a geographically distributed cluster where clients directly send requests to speci c nodes. It is also possible to emulate WAN delays between various cluster nodes and the clients by employing a delayrouter [16] between the nodes. However, as WAN delays do not a ect the results presented in Section 4.3, I did not emulate these delays. The requests were generated by a client program based on the S client architecture [16] The program generates HTTP requests as fast as the Web server can handle them. ....

[Article contains additional citation context not shown here]

G. Banga and P. Druschel. Measuring the capacity of a Web server under realistic loads. World Wide Web Journal (Special Issue on World Wide Web Characterization and Performance Evaluation), 1999.

....the case for reordered in scout. Finally, our profile does not identify the use of data embedded in the code section. This may cause faults on access to code pages containing data. Reducing the number of bytes transfered has benefits other than reducing the startup latency of web applications. Banga Druschel 99, Bradford Crovella 98] showed that having a large number of open connections on a web server can cause serious performance degradation. By reducing the amount of work that the server has to perform, our techniques can reduce the duration of each connection hence reducing the load on servers ....

Banga, G. and Druschel, P. Measuring the Capacity of a Web Server Under Realistic Loads. World Wide Web Journal, May 1999. to appear.

No context found.

G. Banga and P. Druschel. Measuring the capacity of a Web server under realistic loads. World Wide Web Journal (Special Issue on World Wide Web Characterization and Performance Evaluation), 1999.

No context found.

G. Banga and P. Druschel. Measuring the capacity of a Web server under realistic loads. World Wide Web Journal (Special Issue on World Wide Web Characterization and Performance Evaluation), 1999.

....to resume an earlier TLS session and thus reuse the result of an earlier RSA computation. Research has suggested that, indeed, session caching helps web server performance [11] Likewise, there has been considerable prior work in performance analysis and benchmarking of conventional web servers [15, 12, 17, 5, 18], performance optimizations of web servers, performance oriented web server design, and operating system support for web servers [13, 22, 6, 7, 21] Apostolopuolos et al. 3] studied the cost of TLS connection setup, RC4 and MD5, and proposed TLS connection setup protocol changes. Our methodology ....

....[14] We have chosen the Apache mod SSL solution due to its wide availability and use, as shown by a March 2001 survey [26] The TLS implementation used in our experiments by mod SSL is the open source OpenSSL 0.9. 5a [19] The HTTPS traffic load was generated using the methodology of Banga et al. [5], with additional support for OpenSSL. As we are interested primarily in studying the CPU performance bottlenecks arising from the use of cryptographic protocols, we needed to guarantee that other potential bottlenecks, such as disk or network throughput, did not cloud our throughput measurements. ....

G. Banga and P. Druschel. Measuring the capacity of a Web server under realistic loads. World Wide Web Journal (Special Issue on World Wide Web Characterization and Performance Evaluation), 1999.

....a number of client machines, connected to a cluster server. The cluster nodes are 300MHz Intel Pentium II based PCs with 128MB of memory. All machines are con gured with the FreeBSD 2.2.6 operating system. The requests were generated by a HTTP client program designed for Web server benchmarking [8]. The program generates HTTP requests as fast as the Web server can handle them. Seven 166 MHz Pentium Pro machines con gured with 64MB of memory were used as client machines. The client machines and all cluster nodes are connected via switched 100Mbps Ethernet. The Apache 1.3.3 [3] Web server was ....

G. Banga and P. Druschel. Measuring the capacity of a Web server under realistic loads. World Wide Web Journal (Special Issue on World Wide Web Characterization and Performance Evaluation), 2(1), May 1999.

....requests and distributes them to the back end nodes in the cluster. The second con guration shown in Figure 3 emulates a geographically distributed cluster where clients directly send requests to speci c nodes. The requests were generated by a client program based on the S client architecture [6]. The program generates HTTP requests as fast as the Web server can handle them. Seven 166 MHz Pentium Pro machines were used as client machines. The client machines and all cluster nodes are connected via switched 100Mbps Ethernet. The Apache 1.3.9 [2] Web server was used at the server nodes. ....

G. Banga and P. Druschel. Measuring the capacity of a Web server under realistic loads. World Wide Web Journal (Special Issue on World Wide Web Characterization and Performance Evaluation), 1999.

....compared to LAN environment. As a result, at a given throughput in requests second, a real server handles a significantly larger number of concurrent connections than a server tested under LAN conditions [24] The number of concurrent connections can have a significant impact on server performance [4]. Our next experiment measures the impact of the number of concurrent HTTP connections on our various servers. Persistent connections were used to simulate the effect of long lasting WAN connections in a LAN based testbed. We replay the ECE logs with a 90MB data set size to expose the performance ....

G. Banga and P. Druschel. Measuring the capacity of a Web server under realistic loads. World Wide Web Journal (Special Issue on World Wide Web Characterization and Performance Evaluation), 1999. To appear.

No context found.

G. Banga and P. Druschel. Measuring the capacity of a web server under realistic loads. World Wide Web, 2(1-2):69--83, 1999.

No context found.

Banga, G. and Druschel, P. 1999. Measuring the capacity of a web server under realistic loads. World Wide Web 2, 1-2, 69-83.

No context found.

G. Banga and P. Druschel. Measuring the capacity of a web server under realistic loads. World Wide Web, 2(1-2):69--83, 1999.

No context found.

G. Banga and P. Druschel, "Measuring the capacity of a web server under realistic loads," in WWW, 1999.

No context found.

G. Banga and P. Druschel. Measuring the capacity of a web server under realistic loads. World Wide Web, 2(1-2):69--83, 1999.

No context found.

G. Banga and P. Druschel. Measuring the capacity of a web server under realistic loads. World Wide Web, 2(1-2):69--83, 1999.

No context found.

G. Banga, P. Druschel, Measuring the capacity of a Web server under realistic loads, World Wide Web Journal, 2(1), May 1999.

No context found.

Gaurav Banga and Peter Druschel (1999). "Measuring the Capacity of a Web Server Under Realistic Loads." World Wide Web 2(1-2): 69-83. citeseer.nj.nec.com/banga99measuring.html

No context found.

G. Banga and P. Druschel, "Measuring the Capacity of a Web Server Under Realistic Loads," World Wide Web Journal Special Issue on World Wide Web Characterization and Performance Evaluation, vol. 2, no. 1, May 1999, pp. 69-83.

No context found.

BANGA,G.AND DRUSCHEL, P. 1997. Measuring the capacity of a Web server under realistic loads. World Wide Web J. 2,1,69--83.

First 50 documents

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