Electrical computers and digital processing systems: multicomput – Distributed data processing
Reexamination Certificate
1999-01-25
2001-11-13
Follansbee, John A. (Department: 2154)
Electrical computers and digital processing systems: multicomput
Distributed data processing
C709S226000
Reexamination Certificate
active
06317775
ABSTRACT:
BACKGROUND OF THE INVENTION
The present invention relates to methods and apparatus for simulating a virtual server by distributing connection requests over multiple servers at an Internet site. More specifically, the invention relates to methods and apparatus for monitoring the availability of servers at an Internet site which simulate a virtual server and preferentially sending new connection requests to servers which are available for connections and which are likely to have faster response times.
With the recent explosive growth of the Internet, some Internet sites have experienced a very high demand for their services. Many busy sites require multiple servers to adequately service their demands. It is not uncommon for 20 or 30 servers to be dedicated to a given site. The environment of an exemplary multiple server site is shown in
FIG. 1A
, where a group of servers
112
work together to service requests made to their site. Such servers might be World Wide Web servers, for example.
Normally, a human user of the Internet addresses his or her message to a particular internet site by specifying a particular domain name (for example, www.NameX.com). The servers at that site are all associated with the same domain name. When a connection request is received from a user, one of the servers at the internet site must be selected to receive the packet from the user containing the connection request. The packet is then directed to that physical machine. Currently, selection of a server is accomplished by the Domain Name Service (DNS) server which returns an IP address for the domain name.
For convenience, Internet users typically address messages to a domain name rather than to an IP address. This makes it necessary to convert the domain name to an IP address before the message can be routed to a server. To simplify the procedure for accomplishing this, the Internet protocol known as DNS was developed in 1983 by a group of Internet administrators. Typically, a domain name server is provided at the front end of an Internet site. DNS is implemented so that incoming messages addressed to the site can be converted to the proper IP addresses. DNS simply maps domain names provided by a user attempting to access an internet site to one of the IP address which corresponds to a server that serves the site corresponding to the domain name.
While a multiple server site typically has a single domain name which applies to the entire site and therefore to all of the servers at the site, each separate server has its own globally unique IP address so that the server may be individually accessed. For a given server to receive a message, that message must be directed to the server's IP address.
The DNS server determines which physical machine will service a particular request by the IP address given in response to DNS requests. The user therefore selects the domain name and a particular server IP address is assigned for a server that services that domain. This approach works since the user does not typically care which of the machines at a multiple server site services his or her connection request. It is, however, desirable that each connection request be responded to as quickly as possible.
In sites having multiple servers, more than one server is associated with a given domain name and so there must be some mechanism for distributing incoming messages for the same domain name among individual servers with different IP addresses. For this mechanism, most enterprises with multiple servers for a given site now employ a technique known as “round robin DNS.” Round robin DNS simply cycles the IP addresses of the individual server machines in response to successive requests to access the site by name so that the servers take turns handling connection requests.
For example, in a group of servers
112
shown in
FIG. 1A
, an initial request to access the site might be directed to a server
112
A. The second request would then be directed to a server
112
B; the third request would be directed to a server
112
C; and so on, until all of the servers in group of servers
112
have received a single request for access. Then, the next successive request for access is directed back to server
112
A. The next request thereafter is directed to server
112
B, and so on in a round robin manner.
While this technique does help distribute a site's traffic load over a plurality of servers, it certainly does not guarantee that incoming requests will be routed to the servers best able to handle such requests. The round robin distribution scheme is arbitrary and therefore does not necessarily direct incoming requests to the least loaded server. Round robin DNS, in fact, will often send an incoming request to an already heavily loaded server. That request will therefore not be acted upon as promptly as it might have been had it been routed to a different server that was less heavily loaded.
One reason that round robin DNS routes some incoming connection requests to slower servers is that it does not take into account the varying load capacities of the machines and the varying demands on the resources of the machines by different connection requests. Therefore, a given machine in the round robin rotation may become excessively loaded with connections due to its own lack of capacity or the greater requirements of certain of its connections compared to connections made to other machines in the round robin rotation. The excess loading on an individual machine may negatively effect responsiveness of the machine. In such a case, this will result in excessive time delays by this machine whenever data is requested from it. It would be desirable if machines with slower response times or greater connection loads could be identified and if machines with much faster response times or lesser connection loads could be identified, so that connections could be selectively routed to the more responsive machines and the connection loads could be decreased on more loaded and therefore less responsive machines.
A technique known as “DNS caching” creates another problem for round robin DNS. DNS caching is a procedure which is implemented by certain Internet service providers or other clients who often send requests to a given domain name. By noting the source IP address in the packet headers of the packets which are sent from the servers, a client may learn the IP addresses of the servers associated with that domain name. The client may then cache the IP address so that future connection requests addressed to that domain name can be directed to the specific IP address at the service site. Thus, the client bypasses the DNS and chooses the IP address of the physical machine which will service its connection request directly. While this saves the computational time normally required to look up IP addresses on the Internet, it effectively circumvents the round robin DNS connection allocation system (or any other DNS allocation system) and therefore prevents the load on a set of servers from being distributed according to a planned scheme. Clients learn the IP addresses for individual servers and access them at will.
Thus, what is needed is an improved system and method for distributing the load among a plurality of servers at an Internet site. Additionally, it would be desirable if a system and method were developed for preventing clients from accessing a server directly and bypassing the distribution scheme which is implemented.
SUMMARY OF THE INVENTION
The present invention provides a system and method for distributing connections among a plurality of servers at an Internet site. All connections are made to a single IP address and a local director selects the server from among the plurality of servers which is to receive the connection. Thus, the DNS server is not relied upon to distribute connections, and the connection distribution scheme is not avoided when DNS is bypassed. In one embodiment, a session distribution scheme is implemented such that connections are distributed to the server in the group of servers which has the fewest
Coile Brantley W.
Howes Richard A.
Kersey Edward C.
Tenereillo Peter A.
Cisco Technology Inc.
Follansbee John A.
Van Pelt & Yi LLP
LandOfFree
System for distributing load over multiple servers at an... 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 for distributing load over multiple servers at an..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and System for distributing load over multiple servers at an... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2574178