Electrical computers and digital processing systems: multicomput – Computer network managing – Network resource allocating
Reexamination Certificate
1998-06-04
2001-10-30
Dinh, Dung C. (Department: 2153)
Electrical computers and digital processing systems: multicomput
Computer network managing
Network resource allocating
C709S225000, C710S241000, C710S240000
Reexamination Certificate
active
06311217
ABSTRACT:
BACKGROUND OF THE INVENTION
This invention relates generally to computer systems and networks. More specifically, the invention relates to methods and apparatuses for improving the administration of a cluster of computers.
A computer cluster typically consists of a number of computers that require direct access to one or more resources, such as a shared data storage device. Clusters allow a number of computers or servers to have access to the same services. Simultaneous access to the same services is especially useful to carry out transactions from different points of entry. Every time a transaction occurs the information can be updated on a common database. This ensures that the information will remain consistent since the information is kept on the shared data storage device.
FIG. 1A
is a block diagram of a prior art cluster system
100
. Cluster
100
includes servers
102
and
104
, small computer systems interface (SCSI) bus
106
, and storage device
110
. Cluster
100
is also typically connected to a network
120
through servers
102
and
104
. Servers
102
and
104
are coupled to each other and storage device
110
through SCSI bus
106
.
Normally, a client within network
120
will need to obtain or update information stored on storage device
110
. The client will contact one of the servers
102
or
104
in order to carry out the transaction. However, one or both of the servers may not have access to the storage device
110
.
Access to storage device
110
is dependent upon whether servers
102
and
104
are members of the cluster. Generally, a cluster consists of an owner and zero or more members. The owner of the cluster determines whether another computer can have access to a resource. For example, server
104
may be the owner and server
102
may not yet be member of the cluster. In that case, server
102
does not have access to a resource, in this case storage device
110
.
A conventional method of determining ownership is discussed with reference to FIG.
1
B and in conjunction with FIG.
1
A.
FIG. 1B
is a flow chart
140
of a conventional method of cluster administration. The flow chart
140
begins at block
150
and proceeds to block
152
. In block
152
, server
102
attempts to join the cluster. Server
102
initially attempts to communicate with server
104
through network
120
in order to join the cluster as a member. Server
102
assumes that server
104
is the owner of storage device
110
because server
104
is the only other server connected to storage device
110
.
In block
154
, server
102
determines if the attempt to join the cluster as a member was successful. If it was successful, server
102
proceeds to block
160
and joins the cluster as a member. If the communication of block
152
was not successful, server
102
assumes that server
104
is not the owner of storage device
10
.
Proceeding to block
156
, server
102
attempts to gain control of SCSI bus
106
. In the prior art system, control of the SCSI bus equates to control over the storage device. Server
102
then determines if its attempt to gain control over SCSI bus
106
is uncontested in block
158
. If server
104
was actually the owner of the storage device, server
104
would eventually attempt to regain control over the SCSI bus
106
and the storage device
110
.
If server
104
regains control over the SCSI bus, server
102
returns to block
152
and tries to attempt to join as a member through network
120
since it is clear that server
104
is the owner. On the other hand, if no other server has regained control over the SCSI bus
106
and the storage device
110
, server
102
joins the cluster as the owner of the SCSI bus
106
and the storage device
110
in block
159
. When server
102
has joined the cluster as a member in block
160
, or as the owner in block
159
, the processing ends in block
162
.
The conventional method and system of cluster administration have many flaws. For example, conventional cluster systems are generally limited to only those servers or computers that can directly communicate with a common resource. The conventional software system's is typically incapable of handling more than two servers per resource. The limitation of two computers severely limits the versatility and reliability of the cluster. Should one of the servers fail, only one server would be left to provide access to the resource to the network. Further, having only two points of access to the resource limits the frequency of transactions that may be performed with the resource. Thus, the operation of the network may be hindered due to the latencies involved in transactions with the resource.
A cluster system that includes more than two access points would provide greater versatility. Additionally, a cluster system with an independent entry system would increase reliability and decrease transactional arbitration requirements in order to gain access to a storage device.
SUMMARY OF THE INVENTION
The present invention provides a cluster administration system that is capable of handling a cluster having one or more computing devices. The number of computing devices that may be included in a cluster is limited only by practical considerations.
In one embodiment, a cluster administration system includes a cluster of computing devices, one of the computing devices being an owner. The cluster further includes a storage device. Direct access to the storage device by the computing devices is controlled by the owner of the cluster. The cluster administration system also includes an arbiter. The arbiter and the cluster are in communication with each other and a network, the cluster providing the network with access to the storage device. The arbiter controls the admission of new computing devices to the cluster when the owner of the cluster is incapable of admitting the new computing device.
In another embodiment, the arbiter determines which of the computing devices in the cluster is designated as the owner of the cluster. The arbiter can assign a new owner if the current owner looses communication with the arbiter. Also, if the current owner is incapable of admitting a new computing device to the cluster, the arbiter is configured to admit the new computing device as the new owner of the cluster, in another embodiment.
In a further embodiment, the arbiter is implemented on an independent computing device that is independent of the cluster of computing devices. The independent computing device is in communication with the cluster and the network, such that a new computing device desiring to enter the cluster can communicate with the independent computing device through the network.
In yet another embodiment, the arbiter is distributively implemented on an independent cluster of computing devices that is independent of the cluster of computing devices. The independent cluster of computing devices is in communication with the cluster and the network. A first independent computing device of the independent cluster primarily acts as the arbiter for the cluster of computing devices. If the first independent computing device is incapable of primarily acting as the arbiter, a second independent computing device of the independent cluster primarily acts as the arbiter for the cluster of computing devices. In an alternative embodiment, two clusters of computing devices act as arbiters for each other. Any number of clusters may act as arbiters for each other.
Independent arbitration removes some of the burden of cluster administration from the owner. Reliability of the administration of the cluster also is increased. The computing devices of the cluster, and computing devices desiring to enter the cluster need only be able to communicate with the arbiter. When new computing devices are added to a cluster there is no contention for ownership because the arbiter determines which of the computing devices is the owners. Reducing contentions provides better efficiency and reliability.
Having the arbiter outside the cluster provides greater reliability. The arbiter is n
Ehlinger Early David
Fletcher Mark F.
Akin Gump Strauss, Hauer & Feld. L.L.P.
Compaq Computer Corporation
Dinh Dung C.
Trim N
LandOfFree
Method and apparatus for improved cluster administration 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 apparatus for improved cluster administration, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for improved cluster administration will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2593605