Error detection/correction and fault detection/recovery – Data processing system error or fault handling – Reliability and availability
Reexamination Certificate
2000-01-10
2004-06-29
Beausoliel, Robert (Department: 2184)
Error detection/correction and fault detection/recovery
Data processing system error or fault handling
Reliability and availability
C714S011000, C714S012000, C714S013000
Reexamination Certificate
active
06757836
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to computer systems and, more particularly, to improved methods and apparatus for maintaining full connectivity in clustered computer systems.
2. Description of the Related Art
In contrast to single mainframe computing models of the past, more distributed computing models have recently evolved. One such distributed computing model is known as a clustered computing system.
FIG. 1
illustrates an exemplary clustered computing system
100
including computing nodes (nodes) A, B and C, storage devices (e.g., storage disks
102
-
104
), and other devices
106
-
110
representing other devices such as scanners, printers, digital cameras, etc. For example, each of the nodes A, B and C can be a computer with its own processor and memory. The collection of nodes A, B and C, storage disks
102
-
104
, and other devices
106
-
110
make up the clustered computing system
100
.
Typically, the nodes in a cluster are coupled together through a “private” interconnect with redundant pathways. As shown in
FIG. 1
, nodes A, B and C are coupled together through private communication channels
112
and
114
. For example, the private communication channels
112
and
114
can adhere to Ethernet, ATM, or Scalable Coherent Interface (SCI) standards. A client
116
can communicate with the clustered computing system
100
via a network
118
(e.g., public network) using a variety of protocols such as Transmission Control Protocol (TCP), User Datagram Protocol (UDP), etc. From the point of view of the client
116
, the clustered computing system
100
is a single entity that can provide the client
116
with a variety of computer-implemented services, e.g., web-hosting, transaction processing, etc. In other words, the client
116
is not aware of which particular node(s) of the clustered computing system
100
is (are) providing it services.
The clustered computing system
100
provides a scalable and cost-efficient model where off-the-shelf computers can be used as nodes. The nodes in the clustered computing system
100
cooperate with each other to provide a distributed computing model that is transparent to users, e.g., the client
116
. In addition, in comparison with single mainframe computing models, the clustered computing system
100
provides improved fault tolerance. For example, in case of a node failure within the clustered computing system
100
, other nodes can take over to perform the services normally performed by the node that has failed.
Typically, nodes in the clustered computing system
100
send each other “responsive” (often referred to as “heart beat” or activation) signals over the private communication channels
112
and
114
. The responsive signals indicate whether nodes are active and responsive to other nodes in the clustered computing system
100
. Accordingly, these responsive signals are periodically sent by each of the nodes so that if a node does not receive the responsive signal from another node within a certain amount a time, a node failure can be suspected. For example, in the clustered computing system
100
, if nodes A and B do not receive a signal from node C within an allotted amount of time, nodes A and B can suspect that node C has failed. In this case, if nodes A and B are still responsive to each other, a two-node sub-cluster (AB) results. From the perspective of the sub-cluster (AB), node C can be referred to as a “non-responsive” node. If node C has really failed then it would be desirable for the two-node sub-cluster (AB) to take over services from node C. However, if node C has not really failed, taking over the services performed by node C could have dire consequences. For example, if node C is performing write operations to the disk
104
and node B takes over is the same write operations while node C is still operational, data corruption can result.
It should be noted that the fact that nodes A and B have not received responsive signals from node C does not necessarily mean that node C is not operational with respect to the services that are provided by node C. Other events can account for why responsive signals for node C have not been received by nodes A and B. For example, the private communication channels
112
and
114
may have failed. It is also possible that node C's program for sending responsive signals may have failed but node C is fully operational with respect to the services that it provides. Thus, it is possible for the clustered computing system
100
to get divided into two or more functional sub-clusters wherein the sub-clusters are not responsive to each other. This situation can be referred to as a “partition in space” or “split brain” where the cluster no longer behaves as a single cohesive entity. In this and other situations, when the clustered computing system no longer behaves as a single cohesive entity, it can be said that the “integrity” of the system has been compromised.
In addition to partitions in space, there are other potential problems that need to be addressed in managing the operation of clustered computing systems. For example, another potential problem associated with operating clustered computing systems is referred to as a “partition in time” or “amnesia.” As is known to those skilled in the art, partitions in time can occur when a clustered computing system is operated with cluster configurations that vary over time.
Another problem that can affect clustered computing systems is loss of full connectivity. It is common that nodes of a clustered computing system be connected to every other node in the clustered computing system. Some software that is run on clustered computing systems even assumes, and thus requires, that the clustered computing systems have full connectivity. Hence, problems result when such clustered computing systems lose full connectivity. The loss of full connectivity means that the clustered computing system has incomplete (or partial) connectivity. Normally, the incomplete connectivity is caused by failure of an interconnect that couples nodes together. The loss of full connectivity can cause software to crash or “hang”. Accordingly, conventional approaches, such as described in U.S. Pat. No. 6,002,851, maintain full connectivity through use of sophisticated, centralized approaches that attempt to determine which nodes to shut down so that the remaining active nodes of the clustered computing system are fully connected. The disadvantage of this conventional approach is that in order to obtain an optimal solution it is overly complex. As a result of the complexity, the software implementing the optimal solution is complex and lengthy and thus prone to “bugs” (i.e., defects).
In view of the foregoing, there is a need for improved techniques to maintain full connectivity in clustered computer systems.
SUMMARY OF THE INVENTION
Broadly speaking, the invention pertains to techniques for maintaining full connectivity in a clustered computing system. The improved techniques allow for detection of one or more disconnections that cause a loss of full connectivity and then resolution of the disconnections by shutting down one or more appropriate nodes of the clustered computing system to regain full connectivity. As a result, the clustered computing system can effectively maintain full connectivity as is often required by software running on the nodes of the clustered computing system.
The invention can be implemented in numerous ways, including a method, a system, an apparatus, or a computer readable medium. Several embodiments of the invention are discussed below.
As a method for monitoring full connectivity in a clustered computing system having more than two nodes, one embodiment of the invention includes the acts of: detecting loss of full connectivity in the clustered computing system; determining, at each of the nodes, which one or more of the nodes of the clustered computing system should be shut down to regain full connectivity in the clustered computing system; and shutting down the one or more
Kumar Krishna
Murphy Declan J.
Beausoliel Robert
Beyer Weaver & Thomas LLP.
Maskulinski Michael
Sun Microsystems Inc.
LandOfFree
Method and apparatus for resolving partial connectivity 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 Method and apparatus for resolving partial connectivity in a..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for resolving partial connectivity in a... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3356510