Joe Touch, John Heidemann, and Katia Obraczka, "Analysis of HTTP performance." http://www.isi.edu/lsam/publications/httpperf /index.html, August 1996.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... byte tra c may not be nearly as signi cant in such an environment [28] Furthermore, the improvement in latency obtained through the use of persistent connections decreases as the user connection speed decreases, and the main component of round trip time delay becomes transmission time on the link [180]. In practice, popular browsers still use multiple parallel connections to the same server, even though they implement persistent connections. This behavior is discouraged by the HTTP 1.1 standard, which recommends using no more than 2 such connections [64] 62 The interaction of the modi ed ....

Touch J., Heidemann J., Obraczka K., Analysis of HTTP Performance, USC-ISI Technical Report 98-463, December 1998.

....simulation times for large experiments than the fine grained network oriented simulator ns, and to be able to incorporate code that optionally estimates the e#ects of prefetching. The code for HTTP over TCP, and especially for slow start e#ects, is based significantly on the models shown in [THO96, HOT97, KR00] Heidemann et al. HOT97] claim their model provides guidance for wide area and low bandwidth network conditions, but may be somewhat inaccurate when applied to LANs. Therefore, we anticipate 149 similar performance, which will be shown in Section 8.4.2. Fortunately, the area of ....

Joe Touch, John Heidemann, and Katia Obraczka. Analysis of HTTP performance. Available at http://www.isi.edu/lsam/publications/http-perf/, August 1996.

....et al. [15] examine improving HTTP latency. They modify HTTP 1.0 to avoid separate TCP connection for each file requested to access web page. They propose GETALL , GETLIST methods to make pipeline requests which eventually reduce latency for a page that has multiple images. Touch et al. [16] also analyze the performance of HTTP. They discuss the performance effects of using per transaction TCP connection for HTTP access, and proposed optimizations of avoiding per transaction re connection and TCP slow start restart overheads. Transaction TCP provides transaction oriented service ....

....re connection and TCP slow start restart overheads. Transaction TCP provides transaction oriented service over TCP via extensions to the TCP protocol. HTTP performance can be improved by using persistent connections, in which a single connection can access multiple file [14] Touch et al. [16] suggest that the persistent connection optimizations do not substantially affect Web access for the vast majority of users. Most users see end to end latencies of about 250 ms and use modem lines. Bandwidths over 200 Kbps are required to provide user noticeable performance improvements. This ....

Touch, J., J. Heidemann, K. Obraczka, "Analysis of HTTP Performance," USC/Information Sciences Institute, June, 1996.

....for small data transfers (like the Web s requests and responses) Nagle algorithm, which tries to avoid the overhead of small packets can introduce delays to Web requests . Multiple connections are often used for physically co located transfers These and other problems have been discussed in [10, 13, 12, 3, 6, 9, 4]. Many of the problems mentioned above are inherent to the HTTP 1.0 protocol and attempts have been made to fix them in HTTP 1.1. All of those changes concentrated on the interactions between HTTP and TCP protocols, trying to optimize how TCP is used within the HTTP protocol. There is also an ....

J Touch, J. Heidemann, and K. Obraczka, "Analysis of HTTP Performance," USC/Information Sciences Institute, June 1996. http://www.isi.edu/lsam/publications/http-perf/

....cases. For instance, HTTP 1. 0 [39] the first version of the Hypertext Transfer Protocol, opens a new TCP connection for the retrieval of each object in an HTML (Hypertext Markup Language) document: a page with three images, for example, involves the establishment of four distinct TCP connections [40, 41]. HTTP 1.1 [42] solves this problem by introducing persistent connections: the first TCP connection that is used to fetch the FIGURE 2. A wired and a mobile host access the same web server. Radio tower 106 ms 33 ms Web server Wired host Mobile host 1 ms 33 ms 1 ms IEEE Communications ....

J. Touch, J. Heidemann, and K. Obraczka, "Analysis of HTTP Performance," ISI Research Report ISI/RR-98-463, Aug. 1998.

....is quite small. The theoretical rates are estimated along with the percentage of bandwidth lost due to slow start are shown in Table 3. This table shows that for 10 GB files the speedup is only . 3 , but for 100 MB files the speedup would be 23 (speedup is computed by the formula defined in [THO96]) However one can envision a scenario where one might be using http to transfer lots of 100 MB files over a dedicated channel where congestion is not an issue. In this case, the current slow start mechanism is very inefficient. Methods for saving and reusing the previous window size, such as that ....

Touch, J., Heidemann, J., Obraczka, K., "Analysis of HTTP Performance", USC / Information Sciences Institute white paper, http://www.isi.edu/lsam/publications/http-perf/

....significant e#ect on the connection time. Average connection time from a client via 33.6 K modem is two times longer than that from a client via switched Ethernet. In both switched Ethernet and modem cases, connection time is at least a quarter of the total elapsed time. Contrary to Touch s result [5], our study suggests that even for modem users, migrating client to HTTP 1.1 browsers can achieve a significant response time improvement. Simple proxy caching does not always reduce response time. For switched Ethernet users, connection time with the proxy is 1.8 times longer than with ....

Joe Touch, John Heidemann, and Katia Obraczka. Analysis of HTTP Performance. August, 1996 URL:http://www.isis.edu/lsam/publications/http perf/ 6

.... [MAWM98] Nau98] NLN98] Pad95] PM96] Par96] Pet98] PR94] PK96] Pit97] Pit98] RCG98] RSGR00] RF98b] RF98a] RV98] RBR98] RS98] RW98] RS99] RID00] RH98] SKS98] SSV96] SSV97] Sch96] SSV98] SSV99] SAYZ99] SW97] TM00] TG97a] TG97b] TVDS98] THVK98] Tew98] GR96] GR98] THO96] Tou98] Tur98] VR98] VL98] VdJM98] Vix98] VAK 98] Wac96] WC96] Wan99] WMS98b] Wes95a] Wes95b] WC97b] WC97a] WAS 96] Wil98] WS97] WM99b] WM99a] WVS 99] WAB 96] Woo96] WWB96] WA97] Wor94] YFVI00] YS98] ZFJ97] ZMN 98] 1 ....

Joe Touch, John Heidemann, and Katia Obraczka. Analysis of HTTP performance. Available at http://www.isi.edu/lsam/publications/http-perf/, August 1996.

....tests consisted of repeated requests for files of varying sizes: 1K through 1024K, doubling the size of the file at each increment. I had originally considered only large transfers (32K and larger) since the larger the file, the greater the impact of compression. However, as reported in [77], most HTTP transfers are around 6K in size, so I extended my analysis to include smaller transfers. Two different types of files (data) were considered, text and image. The primary purpose of the HTTP proxy is to handle text files (since compression is used) but since a large portion of HTTP ....

J. Touch, J. Heidemann, and K. Obraczka. Analysis of HTTP Performance. URL: http://www.isi.edu/lsam/publications/http-perf, August 1996.

....breaking it down into its constituent parts is not trivial, people often make decisions about how to improve their web service based on little more than guess work. Work to date, as summarised in Section 6, has focused on improving web server performance or web communication protocol efficiency [4,5,6,7] with little regard for the users perspective. In this paper we make three contributions: we present a simple structured timing model to guide analysis of total delay; we describe a number of lightweight methods for measuring the constituent parts of the total delay; we list a set of ....

....that the multiple TCP connections required by HTTP 1.0 are a significant cause of latency in the Web [5] Spero [6] analyses a typical HTTP transaction with timings and concludes that the interaction between HTTP and TCP is a major cause of delay in the Web. On the other hand work reflected in [7] suggests that the overhead of multiple TCP connections is insignificant for low bandwidth users and that its effect is only apparent when using high speed connections. Edwards and Rees have an interesting paper about the execution of CGI applications in servers [4] through HTTP. HTTP 1.1 [17] ....

Touch, J, J. Heidemann, K. Obraczka. "Analysis of HTTP Performance" http://www.isi.edu/lsam/publications/httpperf /

....Analyses of these performance problems are available [30] analysis and results from a prototype implementation are in [26] Implementation experience and measurements of actual HTTP 1. 1 (RFC 2068) implementations show good results [39] Alternatives have also been explored, for example, T TCP [27]. Persistent HTTP connections have a number of advantages: By opening and closing fewer TCP connections, CPU time is saved, and memory used for TCP protocol control blocks is also saved. HTTP requests and responses can be pipelined on a connection. Pipelining allows a client to make multiple ....

Joe Touch, John Heidemann, and Katia Obraczka. "Analysis of HTTP Performance", <URL: http://www.isi.edu/lsam/publications/http-perf/index.html>, USC/Information Sciences Institute, June 1996.

....to fetch each URL, increasing the load on HTTP servers and causing congestion on the Internet. The use of inline images and other associated data often require a client to make multiple requests of the same server in a short amount of time. Analyses of these performance problems are available [30] [27]; analysis and results from a prototype implementation are in [26] Implementation experience and measurements of actual HTTP 1.1 (RFC 2068) implementations show good results [39] Alternatives have also been explored, for example, T TCP [27] Persistent HTTP connections have a number of ....

....of these performance problems are available [30] 27] analysis and results from a prototype implementation are in [26] Implementation experience and measurements of actual HTTP 1. 1 (RFC 2068) implementations show good results [39] Alternatives have also been explored, for example, T TCP [27]. Persistent HTTP connections have a number of advantages: By opening and closing fewer TCP connections, CPU time is saved, and memory used for TCP protocol control blocks is also saved. HTTP requests and responses can be pipelined on a connection. Pipelining allows a client to make multiple ....

Joe Touch, John Heidemann, and Katia Obraczka. "Analysis of HTTP Performance", <URL: http://www.isi.edu/lsam/publications/http-perf/index.html>, USC/Information Sciences Institute, June 1996.

....to fetch each URL, increasing the load on HTTP servers and causing congestion on the Internet. The use of inline images and other associated data often requires a client to make multiple requests of the same server in a short amount of time. Analyses of these performance problems are available [30][27]; analysis and results from a prototype implementation are in [26] Persistent HTTP connections have a number of advantages: By opening and closing fewer TCP connections, CPU time is saved, and memory used for TCP protocol control blocks is also saved. HTTP requests and responses can be ....

J. Touch, J. Heidemann, K. Obraczka, "Analysis of HTTP Performance", <URL: http://www.isi.edu/lsam/ib/http-perf/>, USC/Information Sciences Institute, June 1996.

.... account for much of the perceived latency [Panmanabhan and Mogul 94] One study by Touch, Heidemann and Obraczka reported that most users see end to end latencies of about 250 ms and concluded that the persistent connections do not substantially affect Web latency for the vast majority of users [Touch et al. 96] Viles and French studied the availability and latency of the Web and suggested shorter client side time out intervals than those used for TCP connection establishment [Viles and French 95] The studies by the NRG group at Virginia Tech have shown that 30 to 50 hit rates can be achieved by ....

....time. In our case, connection time is still 25 of total elapsed time for modem users without a proxy. It is not necessarily the case that the percentage of time saved from using persistent connections for a page via a modem is less than that via high speed Ethernet. Contrary to Touche s result [Touch et al. 96] this result suggests that even for modem users, migrating client browsers to HTTP 1.1 compatible versions can achieve a significant response time improvement. 7. The distribution of the number of unique embedded images in a page is not normal; rather it follows a Random Walk distribution. The ....

Touch, J., Heidemann, J. and Obraczka. K.: "Analysis of HTTP Performance", Information Sciences Institute, University of Southern California, August, 1996. URL: http://www.isi.edu/lsam/publications/http-perf/index.html.

....needed to achieve high performance over a single HTTP connection. Pipelining, or batching, have been successfully used in a number of other systems, notably graphics protocols such as the X Window System [15] or Trestle [16] in its original RPC based implementation. Touch, Heidemann, and Obraczka [5] explore a number of possible changes that might help HTTP behavior, including the sharing of TCP control blocks [19] and Transaction TCP (T TCP) 17] 18] The extended length of deployment of changes to TCP argued against any dependency of HTTP 1.1 on either of these; however, we believe that ....

....taking advantage of knowledge in the application can result in a considerably faster implementation than relying on such a timeout. 4.1. 2 Nagle Interaction We expected, due to experience of one of the authors, that a pipelined implementation of HTTP might encounter the Nagle algorithm [2] [5] in TCP. The Nagle algorithm was introduced in TCP as a means of reducing the number of small TCP segments by delaying their transmission in hopes of further data becoming available, as commonly occurs in telnet or rlogin traffic. As our implementation can generate data asynchronously without ....

Touch, J., J. Heidemann, K. Obraczka, "Analysis of HTTP Performance," USC/Information Sciences Institute, June, 1996.

....Analyses of these performance problems are available [30] analysis and results from a prototype implementation are in [26] Implementation experience and measurements of actual HTTP 1. 1 (RFC 2068) implementations show good results [39] Alternatives have also been explored, for example, T TCP [27]. Persistent HTTP connections have a number of advantages: By opening and closing fewer TCP connections, CPU time is saved, and memory used for TCP protocol control blocks is also saved. HTTP requests and responses can be pipelined on a connection. Pipelining allows a client to make multiple ....

Joe Touch, John Heidemann, and Katia Obraczka. "Analysis of HTTP Performance", <URL: http://www.isi.edu/lsam/publications/http-perf/index.html>, USC/Information Sciences Institute, June 1996.

No context found.

Joe Touch, John Heidemann, and Katia Obraczka, "Analysis of HTTP performance." http://www.isi.edu/lsam/publications/httpperf /index.html, August 1996.

No context found.

Touch, J., J. Heidemann, and K. Obraczka, "Analysis of HTTP Performance". USC/Information Sciences Institute, June, 1996. Available at "http://www.isi.edu/lsam/publications/http-perf/index.html".

No context found.

J. Touch, J. Heidemann, and K. Obraczka. Analysis of http performance. Technical report, Information Sciences Institute at University of South California, 1996.

No context found.

Touch, J., J. Heidemann, K. Obraczka, "Analysis of HTTP Performance," USC/Information Sciences Institute, June, 1996. Available at http://www.isi.edu/lsam/publications/httpperf /index.html.

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