Electrical computers and digital processing systems: multicomput – Distributed data processing – Client/server
Reexamination Certificate
1999-12-29
2004-08-17
Geckil, Mehmet B. (Department: 2142)
Electrical computers and digital processing systems: multicomput
Distributed data processing
Client/server
C709S238000, C718S105000
Reexamination Certificate
active
06779017
ABSTRACT:
FIELD OF THE INVENTION
The present invention deals with the global Internet network and more particularly to those of the Internet Servers of World Wide Web (WWW)sites organized as a cluster or group of servers forming a single entity.
BACKGROUND OF THE INVENTION
The Internet is the world's largest network, and it has become essential in organizations such as government, academia and commercial enterprises. Transactions over the Internet are becoming more common, especially in the commercial arena. The information that organizations have on their traditional or legacy business applications may now be published and made accessible to a wide audience. This access may include a person checking a bank savings account, making a hotel reservation or buying tickets for a concert. Making this information or service available for their customers is a competitive advantage for any organization. However, regardless of the innovation and potential benefits provided by a company's Internet solution, its value is greatly reduced if the information cannot be accessed in a reasonable response time. The load on an Internet site is unlikely to remain constant. The number of accesses on a Web server can increase for several reasons.
1. Most companies add their Web site's address to television, radio and print advertising and to product catalogues and brochures. Therefore, awareness of the Web site grows.
2. As time passes, the Web site gains better coverage in the on-line search engines.
3. Assuming the site is providing useful information or a useful service to customers, repeat visitors should increase.
4. Most Web sites begin simply, with fairly modest content, mostly text, with some images. As the site designers grow in confidence, more resources are allocated, and as Web users in general increase their modem speeds, most sites move towards richer content. Thus, not only do hit rates increase, but the average data transfer per hit also rises.
5. Most sites begin as presence sites providing corporate visibility on the Internet and making information about the company available to potential customers. Most present sites use predominantly static Hyper Text Marked-up Language or HTML pages. Static pages are generated in advance and stored on disk. The server simply reads the page from the disk and sends it to the browser. However, many companies are now moving towards integration applications that allow users of the Web site to directly access information from the company's existing applications. This could include checking the availability of products, querying bank account balances or searching problem databases. These applications require actual processing on the server system to dynamically generate the Web page. This dramatically increases the processing power required in the server.
There are several ways to deal with the growth of an Internet site, like purchasing an initial system that is much too large. This is one way to deal with Web site growth; however, most companies are not willing to invest large sums of money in a system that is much larger than they require, particularly since the benefits that they will gain from the site have yet to be proven. Most prefer to purchase a minimal initial system and upgrade as the site demonstrates its worth to the company. In this realm of solutions, load balancing between multiple servers is very often used. In this case, the load for the overall site is balanced between multiple servers. This allows scaling beyond the maximum performance available from a single system and allows for easy upgrading by simply installing additional servers and reconfiguring the cluster to use the additional servers. This solution can also provide the added benefit of higher server availability. The load-balancing software can automatically allow for the failure of a single server and balance the load between the remaining sites. Because the Internet model allows the distribution of services among different servers, called Internet Servers, it is definitively feasible not to tie an application to one specific server. Instead, the service belongs to a group of servers; so an additional computer can be added or removed when necessary. However, grouping the set of servers in a single entity implies that load balancing is efficiently performed between these servers so as to actually achieve optimum performance. A discussion on this and more on load balancing can be found, for example, in a paper by Dias et al., “A Scalable and Highly Available Web Server”, Digest of Papers, Compcon 1996, Technologies for the Information Superhighway, Forty-first IEEE Computer Society International Conference (Cat. No. 96CB35911), pp. 85-92, February 1996.
Load-balancing products have made their way to the market. IBM's eNetwork Dispatcher (eND) is one of those products now commercially available. It creates the illusion of having just one server by grouping systems together into a cluster that behaves as a single, virtual server. The service provided is no longer tied to a specific server system; so it is possible to add or remove systems from the cluster, or shutdown systems for maintenance, while maintaining continuous service for the clients. The balanced traffic among servers seems for the end users to be a single, virtual server. The site thus appears as a single IP (Internet Protocol) address to the world. All requests are sent to the IP address of the e Network Dispatcher machine, which decides with each client request which server is the best one to accept requests, according to certain dynamically set weights. Network Dispatcher routes the clients' request to the selected server, and then the server responds directly to the client without any further involvement of eND. This makes it possible to have a small bandwidth network for incoming traffic (like Ethernet or token ring) and a large bandwidth network for outgoing traffic (like ATM—Asynchronous Transfer Mode or FDDI—Fiber Distributed Data Interface or Fast Ethernet). It can also detect a failed server and route traffic around it. General information on the way of performing load balancing between multiple servers and an eND product can be found in a ‘Redbook’ by IBM published by the Austin, Tex. center of the International Technical Support Organization (ITSO) and entitled “Load-Balancing Internet Servers” under the reference SG24-4993 on December 1997.
Those products are great for what they have been designed for, i.e., load-balancing, and indeed allow building of a scalable Web site capable of coping with a rapidly growing demand for higher traffic. However, they have also created their own difficulties. Because there are now numerous sophisficated Web servers that allow handling dynamic Web pages, they need to be session-aware for every user accessing their service. Several techniques indeed exist to keep track of the context in which a particular user is accessing a Web server. They are of two kinds:
the contextual data is circulating, back and forth, in the IP packets exchanged between the client and the servers. For example, it can be part of the Web pages themselves.
or the contextual data is kept in the Web server active memory or on disk. This second solution is necessary whenever the amount of data needed to define each session context is too large to be practically transported over the network with each transaction between the client and the servers.
Then, load-balancing products such as eND manage not to randomly dispatch the traffic to the servers of their cluster. They keep track of the user requests which must end up in the same server while a session is active. To achieve this, the usual technique, well known in the art, comprises utilizing the IP address of the client. Then, each transaction coming from the same IP address is dispatched to the same server.
However, this does not fit in the now frequent situations in which the end user and the server are on either side of a proxy, socks or fire-wall. All those devices, part of the Internet, are intended to deal with specific
Lamberton Marc
Levy-Abegnoli Eric
Montagnon Eric
Thubert Pascal
Clay A. Bruce
Geckil Mehmet B.
International Business Machines - Corporation
Walder, Jr. Stephen J.
Yee Duke W.
LandOfFree
Method and system for dispatching client sessions within a... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and system for dispatching client sessions within a..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and system for dispatching client sessions within a... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3274040