< Previous  | TOC | Index | Next  >

Summary of best practices for cluster communications

The following are the recommended best practices for cluster communications to best support proper cluster membership and data protection.

• Properly seed the cluster by requiring all systems, and not just a subset of systems, to be present in the GAB membership before the cluster will automatically seed.

  If every system is not present, manual intervention by the administrator must eliminate the possibility of a split brain condition before manually seeding the cluster.

• Configure multiple independent communication network links between cluster systems.

  Networks should not have a single point of failure, such as a shared hub or ethernet card.

• Low-priority LLT links in clusters with or without I/O fencing is recommended. In clusters without I/O fencing, this is critical.

  ---

    Note   An exception to this is if the cluster uses fencing along with Cluster File Systems (CFS) or Oracle Real Application Clusters (RAC).

  ---

  The reason for this is that low priority links are usually shared public network links. In the case where the main cluster interconnects fail, and the low priority link was the only remaining link, large amounts of data would be moved to the low priority link. This would potentially slow down the public network to unacceptable performance. Without a low priority link configured, membership arbitration would go into effect in this case, and some systems may be taken down, but the remaining systems would continue to run without impact to the public network.

  It is not recommended to have a cluster with CFS or RAC without I/O fencing configured.

• Disable the console-abort sequence

  Most UNIX systems provide a console-abort sequence that enables the administrator to halt and continue the processor. Continuing operations after the processor has stopped may corrupt data and is therefore unsupported by VCS.

  When a system is halted with the abort sequence, it stops producing heartbeats. The other systems in the cluster consider the system failed and take over its services. If the system is later enabled with another console sequence, it continues writing to shared storage as before, even though its applications have been restarted on other systems.

  Symantec recommends disabling the console-abort sequence or creating an alias to force the go command to perform a restart on systems not running I/O fencing.

• Select the smallest possible LUNs for use as coordinator disks. No more than three coordinator disks are needed in any configuration.
• Do not reconnect the cluster interconnect after a network partition without shutting down one side of the split cluster.

  A common example of this happens during testing, where the administrator may disconnect the cluster interconnect and create a network partition. Depending on when the interconnect cables are reconnected, unexpected behavior can occur.