Cluster repository detection performance improvement by failed node (17-Dec-2009)

Cluster repository detection performance improvement by failed node

Cluster multi-

sharing disk resources. This basically forms a distributed operating system environment. In a cluster, multiple server machines cooperate to provide a set of services or resources to clients. This ensures that there is continuous availability of applications, along with their data residing in the disks.

Typically, a cluster comprises of large number of nodes that are interconnected through ethernet connection. These nodes have mutiple disks connected to them. These disks can be shared across nodes as well.

In the cluster architecture that is being discussed here, information about the cluster including the node names, disks, ODM (Object Data Manager) information etc. will be located in the central shared disk repository which can be accessed by any node in the cluster.

The repository disk could be one among the possible thousands of disks that the cluster nodes shares. So when a node goes down and is coming up it needs to search through all of the thousands of disks to check which is its repository thereby resulting in perfomance issue.

Figure 1: Cluster organization and interaction with Repository.

processing system is a group of loosely coupled machines networked together,

1

The repository disk header has magic number which indicates that the disk is unique with other disks. Searching the repository disk for a node that is rebooted becomes laborious if there are many disks and this causes a significant performance impact. This articl...