Electrical computers and digital processing systems: multicomput – Computer network managing – Computer network monitoring
Reexamination Certificate
1997-11-07
2001-12-11
Harrell, Robert B. (Department: 2152)
Electrical computers and digital processing systems: multicomput
Computer network managing
Computer network monitoring
C709S217000, C709S226000, C709S249000, C709S203000, C370S428000, C370S401000
Reexamination Certificate
active
06330602
ABSTRACT:
FIELD OF THE INVENTION
The invention relates generally to distributed server systems which are used widely in the field of telecommunications for information sharing. In particular, it is directed to a distributed server system which is scaleable and realizes high server performances. The distributed server system of the invention includes an intermediary node between a client and a pool of servers.
BACKGROUND OF THE INVENTION
The last few years have observed a phenomenal growth in web (short for World-Wide Web or WWW, or Internet) usage. This growth has demonstrated the value of wide-area information-sharing but, at the same time, caused a significant research interest in improving the performance of web systems. Recent studies show that the web consumes more Internet bandwidth than any other application.
At a macro level, a web system consists of three components: (i) client, (ii) communication protocol, and (iii) server. Efforts are being made at each component to enhance the performance of the overall web systems.
At the client end, supports are provided to improve response time by the following features: memory cache, disk cache, allowing multiple simultaneous sessions and introducing a proxy server for another level of caching.
The communication protocol between the client and the server is HTTP (Hypertext Transfer Protocol) which always assumes the existence of a reliable path layer underneath the client and server. TCP/IP (Transmission Control Protocol/Internet Protocol) provides reliable data transmission using window flow control techniques. HTTP therefore runs on top of the TCP/IP layer. Asynchronous Transfer Mode (ATM) is another transmission technique to handle broadband multimedia traffic. It continues to grow steadily in the communication world. High-speed ATM switches are available in the commercial market. The co-existence of the Internet and large-scale ATM networks is expected in the near future. ATM can provide wide-area virtual circuits, thus facilitating geographical distribution of web servers.
The HTTP has also undergone changes for performance improvement. It has been reported that multiple TCP sessions per HTTP transaction is a major cause of performance bottleneck. The introduction of a “keep alive” header allows sessions to be kept open and used for multiple HTTP request/response activities.
The web servers have also undergone improvements: first-generation servers handled 20 transactions/sec. based on one process per transaction. The major overhead was due to a large process fork time for a new transaction. This was avoided by pre-forking multiple processes and using a dispatcher to distribute transactions among them to achieve server performance of 100 transactions/sec. The “keep alive” HTTP feature, along with the multi-threaded architecture of only one process, allows the server to handle more than 250 transactions/sec. However, the current trend indicates that the popular sites will incur a significantly higher number of server transactions per second in the near future. This requires more powerful web servers which may be developed by improving different components of a web server (e.g., CPU speed, disk performance, file system performance, performance of TCP/IP, server software architecture etc.). Alternatively, multiple servers can be used to handle high rate of server transactions.
The multiple server approach has two immediate advantages: if a server fails, the session can be handled from other servers; also the total cost of multiple servers can be less than the cost of one server with the equivalent performance. It is therefore foreseen that multiple server systems will be in great demand to accommodate an ever-increasing number of user transactions.
Different architectures for multiple server systems are currently in use and are described briefly here. The use of a Domain Name System (DNS) server to distribute traffic among multiple servers was investigated at NCSA at the University of Illinois.
FIG. 1
shows the DNS system. When a client
10
wishes to communicate with a server
12
, at first it contacts the DNS
14
, from which it obtains the IP address of the desired server. The client then uses this IP address to communicate with the server. All clients perform the same process unless they already have the IP addresses of servers with which they want to connect. When there are a plurality of servers which hold identical information, the DNS rotates in a round robin manner through a pool of these identical servers which are alternatively mapped to the alias of the hostname of one server. This approach has provided some success in distributing the server load, however, it could not balance the load among servers. Another problem with this approach is that, once the IP address resolution is cached in the local memory, the client may never contact DNS.
Another system uses the HTTP level redirection capability to move a transaction among multiple servers.
FIG. 1
also shows this mechanism. When a server
12
finds that it is impossible to handle any extra traffic, it can redirect a transaction to another preselected server
18
and hence distribute the load. HTTP redirection is a common technique used for WWW load distribution. The implementation maybe simple and straightforward, but the redirection requires a round trip delay between the client and server before the transaction is redirected to a different server. Moreover, if the first server is already very busy, the response delay will be even greater.
FIG. 2
shows another known system which switches the load based on the client IP address. Each client
20
goes to an intermediate device
22
which examines the originating IP address and decides where to forward the traffic among multiple servers
24
. IP address hashing is one of the possible mechanisms to determine the server to which the traffic will be directed. This technique, however, lacks the dynamic control of user accesses. Moreover, the IP address spaces are partitioned into five different classes. Care should fherefore be taken in designing good hashing function.
HTTP is a stateless protocol. A web server obtains everything it needs to know about a request from the HTTP request itself. After the request is serviced, the server can forget the transaction. Thus, each request in HTTP is disjointed. If all the servers are identical (or see the same file system using a distributed file system), the server from which the request is served is of little relevance to the client. The choice of a physical server itself is immaterial to the transactions. An HTTP transaction is an aggregation of one or more TCP sessions. Based on this principle, different TCP sessions can be allocated to different servers without the knowledge of whether or not all the TCP sessions belong to the same HTTP transaction. The present invention realizes this TCP-based switching by the use of an intermediary entity called a depot to perform these functions of session allocation. Thus, the TCP-based server switching allows a nice granularity for load balancing among multiple servers. It is also envisaged that this concept of forwarding different sessions to different servers can be applied to similar multi-server architectures of telecommunications networks.
OBJECTS OF THE INVENTION
It is an object of the invention to provide a server system which has a high performance.
It is another object of the invention to provide a communications architecture which allows the construction of a scaleable and high performance server system.
It is a further object of the invention to provide a method of and an apparatus for efficiently managing the resources of a multiple server system.
It is still another object of the invention to provide a method of and an apparatus for efficiently managing the resources of a multiple server system by the use of an intermediate entity.
It is yet a further object of the invention to provide a method of and an apparatus for efficiently managing the resources of a multiple server system by the use of more than one instance of the inte
Chapman Alan S.J.
Law Eddie Ka Lun
Nandy Biswajit
Harrell Robert B.
Nortel Networks Limited
Willett Stephan
LandOfFree
Scaleable web server and method of efficiently managing... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Scaleable web server and method of efficiently managing..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Scaleable web server and method of efficiently managing... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2590330