Electrical computers and digital processing systems: multicomput – Computer-to-computer data routing
Reexamination Certificate
2000-03-24
2004-11-16
Fang, Paul H. (Department: 2141)
Electrical computers and digital processing systems: multicomput
Computer-to-computer data routing
C709S241000
Reexamination Certificate
active
06820133
ABSTRACT:
FIELD OF THE INVENTION
The invention relates to communication between web clients, such as browsers, and web servers, where the communication is carried at least in part over the Internet or any other network layer in which performance criteria such as cost, bandwidth usage and communication latency are of concern.
BACKGROUND OF THE INVENTION
Today, improving the performance of Internet communication is a major technological and commercial concern. Investment in improving the Internet network infrastructure is estimated to become a $1.3 trillion dollar industry by 2003 (Source: Nortel Industries press release Jan. 31, 2000). It has been estimated that many web users will not tolerate a delay in downloading a web page of more than about 8 seconds, and that the current value of e-commerce sales at risk because of slow download speeds is $4.35 billion per year. (Source: Zona Research report “The need for speed”, abstract available http://www.zonaresearch.com/info/press/99-jun30.htm). In this climate there is pressing demand for ways to improve web performance, and no simple or obvious techniques are overlooked.
Standard Internet Protocols: HTTP and TCP
Two standard protocols used on Internet links, HTTP and TCP, impose a significant limitation Internet communication speed. HyperText Transport Protocol (HTTP) is the application-level network protocol used when a client requests web content from a web server, and used by the web server when it responds to such requests. Modern network communication is layered, which means that higher-level protocols build on top of lower-level protocols (which in turn may build on other protocols). HTTP is a high-level protocol which includes commands to request content, respond with content, negotiate the form in which content is sent, and so forth. It is generally carried over the lower-level protocol Transmission Control Protocol (TCP). TCP enables reliable end-to-end connectivity between two locations in the Internet, but does not interpret the content sent between these two locations in any way: it just carries a stream of bytes. TCP in turn is generally carried over the Internet Protocol (IP), which is a packet-oriented protocol that does not guarantee reliable delivery.
It is well-known that HTTP and TCP are far from optimal protocols for Internet communication. TCP was developed and deployed well before HTTP was invented, and was designed for bulk bi-directional data transfer. HTTP is characterized by short request messages and moderate-length response transactions, and very bursty traffic. That TCP is not an optimal protocol for carrying HTTP is extensively documented. The paper “Modeling the Performance of HTTP Over Several Transport Protocols” in IEEE/ACM Transactions on Networking, vol. 5, number 5, October 1997, by Heidemann, Obraczka, and Touch, is representative of research addressing these issues. They claim, for instance:
These mismatches between the needs of HTTP and the services provided by TCP contribute to increased latency for most web users. Fundamentally, TCP is optimized for large-scale bulk data transport, while HTTP often needs a light-weight, request-response protocol.
The mismatches referred to here relate to a number of technical features of TCP, including those known as “three way handshake”, “slow-start congestion avoidance”, and “TIME_WAIT tear-down delays”. The Heidemann, Obraczka and Touch paper discusses several such improved protocols, such as Transaction TCP (T/TCP) and Asynchronous Reliable Delivery Protocol (ARDP). Other defects in TCP as it relates to HTTP include the flow-control algorithm being used, which can lead to unnecessary traffic and delays in the event of noise or error on the network.
HTTP itself is an evolving, improving protocol, but it has recognized performance deficiencies even aside from the interrelationship with TCP. The PhD Dissertation “Addressing the Challenges of Web Data Transport” by V. N. Padmanabhan (Computer Science Division, University of California at Berkeley, USA; Also published as Technical Report UCB/CSD-98-1016 September 1998) discusses some of these. As an example, it explains how HTTP Version 1.0 (still in wide use today) requires a client to send one request at a time over a given connection, waiting for the response to arrive completely before continuing, and at considerable performance cost (as the dissertation proves).
The problem is not that protocols for Internet communication that are better than HTTP and TCP do not exist or are not available. The problem is that HTTP and TCP are standards—widely accepted and widely deployed. Indeed, this is necessarily so, since communication over a shared network such as the Internet requires all users to use the same protocol. Thus, even when problems with existing protocols are noticed and improved protocols developed, it often takes a long time before such improvements become widely deployed. The delays are particularly long before improvements reach the public Internet infrastructure. In part, this delay is simply because costs are always large when significant software upgrades are needed. But in the case of protocol upgrades the costs and delays are even larger because no one can upgrade unilaterally: both ends of a network conversation must be using the same version of the same protocol. In the case of the Internet, some protocol changes require a community-wide coordinated update. For an example of such delays, consider that the problem with HTTP Version 1.0 cited above from Padmanabhan's dissertation was corrected in the next version of the protocol, where a feature known as “pipelining” was defined. Even several years after this improvement was first suggested there are very few web browsers that adopt it. Similarly, all proposals to replace TCP have languished, and today all major web browsers and web servers support HTTP over TCP only.
One prior art approach to improving Internet performance, without altering the standard protocols, is web caching. A similar approach is content distribution (CD). A Content Distribution (CD) network is a collection of specialized nodes or devices, placed in a larger network such the Internet at chosen locations such as in the offices of Internet Service Providers (ISPs). These nodes store certain web content on behalf of the content distributors' customers. Such stores are sometimes called caches, mirrors, or repeaters.
A Content Distribution service includes a redirection or interception service. When a web user (using a client such as a browser) requests content from a site, and the content is known or suspected to be cached at one or more CD nodes, the request is directed (or comes to be redirected) to some CD node that is “close” to the user. The notion of closeness is a measure of communications performance, and in particular can use such metrics as bandwidth capacity, bandwidth cost, latency, security, administrative boundaries, administrative convenience, and current congestion on various network paths. The technologies for choosing a close CD node and then directing requests to the chosen node are varied, but the field is still new and there is still considerable ongoing innovation.
An alternative to redirection of the type just discussed is interception, in which a node is placed in the network path from the client in such a way that it gets to see all web traffic from the client. A web proxy or other specialized device such as a router, for instance located at the client's ISP, can be used for this purpose. In this case, the node intercepts all traffic and if it sees a request for content it has cached (or can readily fetch from a nearby cache) then it can return the content immediately, but otherwise it relays the traffic to its destination unchanged. The use of a proxy may be under the client's control (e.g., if the client must be configured to use a proxy), or be “transparent” if the client needs no such configuration.
The advantages of Content Distribution are the possibility of serving traffic to the user from a close CD node, thus getting the respo
Grove Adam J.
Kharitonov Michael
Tumarkin Alexei
Fang Paul H.
Netli, Inc.
Nguyen Quang
White & Case LLP
LandOfFree
SYSTEM AND METHOD FOR HIGH-PERFORMANCE DELIVERY OF WEB... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with SYSTEM AND METHOD FOR HIGH-PERFORMANCE DELIVERY OF WEB..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and SYSTEM AND METHOD FOR HIGH-PERFORMANCE DELIVERY OF WEB... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3360449