Electrical computers and digital processing systems: multicomput – Computer-to-computer data routing
Reexamination Certificate
1998-02-09
2001-01-16
Geckil, Mehmet B. (Department: 2756)
Electrical computers and digital processing systems: multicomput
Computer-to-computer data routing
C709S205000
Reexamination Certificate
active
06175874
ABSTRACT:
CROSS-REFERENCE TO RELATED APPLICATIONS
This application claims the benefit of Japanese Application No. 9-178062 which is hereby incorporated by reference.
BACKGROUND OF THE INVENTION
The present invention relates to communication mechanisms for computers performing parallel calculation processing in plural packet switched-type networks. In particular, the present application relates to a packet relay control method for realizing load dispersion with high reliability, a packet relay device, and a program memory medium for storing relates program. The present invention also provides a single system image at an address level while distributing request packets, via a network, to appropriate nodes in a cluster.
In a computer network, OSI protocol stack(s) according to ISO, CCITT, or TCP/IP and the like, are used. The protocol stacks act as a transport layer and provide a mechanism which performs communication from a specific process of a transmission host, to a specific process of a reception host. The transport end point is designated by assembling the host designation and service end point designation elements. In TCP/IP, the host designation becomes the IP address, the service end point designation becomes the port number. The receiving computer performs processing which transmits the communication content to the necessary process.
To construct a large scale network server, clusters are used which connect a plurality of personal computers (PC) or workstations (WS). The processing nodes in a cluster have a minimum of one network interface each, and are allotted individual network addresses.
When a client requests a service from a cluster, the client acquires a network address of one node in the cluster and sends a request packet. Normally, the acquisition of a network address requires an inquiry to a name server. The name server performs distributed management of a database using conversion data relating host names, which are a human concept, to network addresses. In a name service, which some name server provides, a cluster appears to be one system, with a plurality of network addresses corresponding to a single host name. For example, the Internet standard DNS (Domain Name System) provides for such name service. The DNS manages the relationship between domain names and IP addresses on the Internet.
When a client makes a request, selecting one address, as each client is different, the request dispersed in nodes within a cluster. When viewed from a client receiving a service, it is desirable for the cluster to appear as though it were a high performance single server. A cluster which appears as one system is said to have a single system image (SSI). By having a SSI, a client can access the cluster in the same way as accessing a single node of a server. One method of implementing an SSI cluster which has plural network interfaces is using round robin DNS.
FIG. 16
is a diagram showing an example of an http cluster consisting of a node
9
C (named) which manages 2 processing nodes
9
A and
9
B (http) and IP addresses. An external network
4
connects user terminals
5
A to
5
M which receive service from the cluster. In general, the nodes
9
A-
9
C provide service for the user terminals
5
A-
5
M. A router (relay device) connects the external network
4
to a cluster network
92
connecting nodes
9
A-
9
C.
Names on the network of the system are registered and referred to using the DNS. In accordance with the DNS, addresses of plural IPs can be registered in connection with one domain name. In the example shown in
FIG. 16
, the nodes
9
A and
9
B appear as one name to the clients of the user terminals
5
A-
5
M. The node “named” (node
9
C) realizes the DNS as a record (A record) of an IP address of a domain name, for example: www.domain.co.jp. The IP address of the network interface of nodes
9
A,
9
B which operate httpd, are registered in a list, such as:
www.domain.co.jp IN A XXX.XXX.XXX.YYY ;; (Address of Node
9
A)
www.domain.co.jp IN A XXX.XXX.XXX.ZZZ ;; (Address of Node
9
B)
From the client's point of view he can access the whole cluster which the name represents. When there is an inquiry from the client for the record A, the node
9
C responds by changing the order. When plural clients access this cluster, because it replies with the IP address of the correct node as an index result, the request is distributed randomly to the httpd nodes
9
A and
9
B.
Normally, because the user who uses a web client has no IP address and is recognized by a domain name, it is sufficient in many cases if the domain name is by one system image. In other words, a round robin DNS at the domain name level is an SSI. However, at the IP address level, two addresses exist, the address of node
9
A and the address of node
9
B. Thus, nodes
9
A and
9
B do not appear as one system. This causes problems such as when a partial failure of a node occurs.
When the node
9
A and the node
9
B provide the same service, consider the case in which node
9
A fails. By an inquiry to the DNS, or a client which has acquired the IP address of the node
9
A (which has failed), because no service is received from the node
9
A, it appears that the httpd service has stopped. Even if the node
9
B puts a connection in motion, insofar as this client does not acquire an address, service is not restored for the client.
SUMMARY OF THE INVENTION
Accordingly, it is an object of the present invention to provide a single system image (SSI) at the address level by distributing a request packet via a network to a node within a cluster.
It is also an object of the present invention to provide a network server which is capable of operating connections.
Additional objects and advantages of the invention will be set forth in part in the description which follows, and, in part, will be obvious from the description, or may be learned by practice of the invention.
REFERENCES:
patent: 5247620 (1993-09-01), Fukuzawa et al.
patent: 5774660 (1998-06-01), Brendel et al.
patent: 5787253 (1998-07-01), McCreery et al.
patent: 5802054 (1998-09-01), Bellenger
patent: 5832219 (1998-11-01), Pettus
patent: 5923654 (1999-07-01), Schnell
patent: 5923660 (1999-07-01), Shemla et al.
patent: 6018526 (2000-01-01), Liu et al.
Imai Yuji
Katsuyama Tsuneo
Kishimoto Mitsuhiro
Fujitsu Limited
Geckil Mehmet B.
Staas & Halsey , LLP
LandOfFree
Packet relay control method packet relay device and program... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Packet relay control method packet relay device and program..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Packet relay control method packet relay device and program... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2497265