Electrical computers and digital processing systems: multicomput – Network computer configuring
Reexamination Certificate
1999-03-10
2001-02-20
Lim, Krisna (Department: 2758)
Electrical computers and digital processing systems: multicomput
Network computer configuring
C709S201000, C709S249000, C370S254000
Reexamination Certificate
active
06192401
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
This invention relates to distributed computer systems, and more particularly to a system and method for dynamically determining cluster membership.
2. Description of the Related Art
As databases and other large-scale software systems grow, the ability of a single computer to handle all of the tasks associated with the database diminishes. Other concerns, such as failure handling and the response time under a large volume of concurrent queries, also increase the number of problems that a single computer must face when running a database program.
There are two basic ways to handling a large-scale software system. One way is to have a single computer with multiple processors running a single operating system as a symmetric multiprocessing system. The other way is to group a number of computers together to form a cluster, a distributed computer system that works together as a single entity to cooperatively provide processing power and mass storage resources. Clustered computers may be in the same room together, or separated by great distances. By forming a distributed computing system into a cluster, the processing load is spread over more than one computer, eliminating single points of failure that could cause a single computer to abort execution. Thus, programs executing on the cluster may ignore a problem with one computer. While each computer usually runs an independent operating system, clusters additionally run clustering software that allows the plurality of computers to process software as a single unit.
Another problem for clusters is how to configure into a cluster or how to reconfigure the cluster after a failure. Initial configuration of the cluster is described in related and co-pending patent application having Ser. No. 08/955,885, entitled “Determining Cluster Membership in a Distributed Computer System”, whose inventors are Hossein Moiin, Ronald Widyono, and Ramin Modiri, filed on Oct. 21, 1997, now U.S. Pat. No. 5,999,712 issued on Dec. 7, 1999. A failure may be hardware and/or software, and the failure may be in a computer node or in a communications network linking the computer nodes. A group of computer nodes that is attempting to reconfigure the cluster will each vote for their preferred membership list for the cluster. If the alternatives have configurations that distinctly differ, an elected membership list for the cluster is often easily determined based on some arbitrarily set selection criteria. In other cases, a quorum of votes from the computer nodes, or a centralized decision-maker, must decide on the cluster membership. A quorum may be defined as the number of votes that have to be cast for a given cluster configuration membership list for that cluster configuration to be selected as the current cluster configuration membership.
One serious situation that must be avoided is the split-brain condition. A split-brain is where two differing subsets of nodes each think that they are the cluster and that the members of the other subset have shut down their clustering software. The split-brain condition leads to data and file corruption, since the two subsets each think that they are the cluster with control of all data and files.
Thus, it can be seen that a primary concern with clusters is to how to determine what configuration is optimum for any given number and coupling of computers after a failure. Considerations such as how many of the available computers should be in the cluster and which computers can freely communicate should be taken into account. It would thus be desirable to have an optimized way to determine membership in the cluster after a failure causes a reconfiguration of the cluster membership.
SUMMARY OF THE INVENTION
The problems outlined above are in large part solved by a system and method for determining cluster membership in a distributed computer system. In one embodiment, the system comprises a plurality of computer nodes coupled through one or more communications networks. These networks may include private and/or public data networks. Each of the computer nodes executes cluster management software that helps determine cluster membership in the distributed computer system. Weighting values assigned to each node are combined to choose an optimal configuration for the cluster. A cluster configuration must be determined upon initiation of a new cluster. Cluster reconfiguration of an existing cluster must also occur if a node joins or leaves the cluster. The most common reason for a node to leave the cluster is by failure, either of the node itself or a communication line coupling the node to the cluster. Basing cluster membership decisions upon weighting factors assigned to each computer node may advantageously increase availability and performance by favoring the most valued (fastest, etc.) nodes in the cluster when nodes must be failed to prevent split-brain configurations.
A method is contemplated, in one embodiment, to determine the membership of nodes in the cluster by assigning a weighting value to each of the nodes. The weighting value may be based upon various factors, such as relative processing power of the node, amount of physical memory, etc. A first subset of the nodes is grouped into a first possible cluster configuration, while a second subset of the nodes is grouped into a second possible cluster configuration. The weighting values of each subset are combined to calculate a first and a second value for the first and second possible cluster configurations, respectively. The membership in the cluster is chosen based on the first and second values. In a further embodiment, the first and second subsets may be but a start to a number of subsets of nodes, each grouped into a possible cluster configuration according to predetermined rules. In this further embodiment, the weighting values are calculated for each possible cluster configuration. The membership in the cluster is chosen based on the weighting values calculated for each possible cluster configuration. This feature may advantageously result in the cluster reconfiguring with an optimized configuration. The method may be implemented in software.
In a further embodiment, the weighting values for the computer nodes are compared to find a node with the maximum weighting value. The maximum weighting value may be adjusted if the maximum weighting value is greater than or equal to the sum of all other weighting values for all other nodes in the current cluster configuration. According to one preferred embodiment, the maximum weighting factor is adjusted to a value below the sum of all other weighting values for all other nodes in the current cluster configuration. This feature may advantageously result in the cluster having an optimized configuration that is less susceptible to single mode failures.
REFERENCES:
patent: 5426674 (1995-06-01), Nemirovsky et al.
patent: 5805785 (1998-09-01), Dias et al.
patent: 5822531 (1998-10-01), Gorczyca et al.
patent: 6014669 (2000-01-01), Slaughter et al.
Chandra et al., “On the Impossibility of Group Membership,” Proceedings of PODC 1996.
Fischer et al., “Impossibility of Distributed Consensus with One Faulty Process,” Journal of the ACM, 32(2):374-382, Apr. 1985.
Sun™ Clusters, A White Paper, Copyright Sun Microsystems, Inc., Oct. 1997.
The Sun Enterprise Cluster Architecture, Technical White Paper, Copyright Sun Microsystems, Inc., Oct. 1997.
Nasypany et al., “Testing Cluster Solutions in an RS/6000 Environment,” Jan. 1997.
Modiri Ramin
Moiin Hossein
Conley Rose & Tayon PC
Kivlin B. Noäl
Lim Krisna
Sun Microsystems Inc.
LandOfFree
System and method for determining cluster membership in 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 System and method for determining cluster membership in a..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and System and method for determining cluster membership in a... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2603534