Impact of Network Adapter Failure in a Cluster

• Microsoft Windows NT Server, Enterprise Edition version 4.0 - Microsoft Cluster Server

SUMMARY

This article contains information regarding network adapter failure within a cluster and different ways the condition may be handled.

MORE INFORMATION

The impact of network adapter failure may vary depending on hardware configuration and the number of network adapters installed. With many different possible network configurations, this article covers the three most common configurations.

Configuration 1

Private Network, Client Network Configured for All Communications:

The Administrator's guide for Microsoft Cluster Server (MSCS) suggests using an isolated network for cluster communications and one adapter for the client network (configured for both client and cluster communications). With this configuration, if the adapter fails for the private network that the cluster nodes use, the nodes may use the client network for their communications and still communicate.

If the client network adapter fails, the IP address resources bound for that adapter will also fail. If the resources cannot be brought online, the group will failover to a surviving node in the cluster. Simply removing the network cable from the client network adapter does not constitute an adapter failure.

If both adapters fail to the point that the cluster nodes cannot communicate, the arbitration process begins for access to the quorum disk. The winner of the process takes control of the cluster and its groups. The loser withdraws from the cluster by shutting down its Cluster service.

Configuration 2

Private Network, Client Network Configured for Clients Only:

MSCS handles this configuration very similarly to configuration 1 noted above. However, if an adapter on the private network fails, there is no other network configured to allow for cluster communications. As a result, the cluster nodes will arbitrate for the quorum disk and the winner will take control of all groups. The loser of the arbitration process will withdraw from the cluster by shutting down its Cluster service.

Configuration 3

No Private Network for Cluster Communications:

While this configuration works, it is further limited. If the adapter fails in one of the nodes, with no other adapter to use as an alternate for cluster communications, the nodes immediately arbitrate for the quorum disk. The winner of the arbitration process takes control of all cluster groups and the losing node shuts down its Cluster service. With no way for the two nodes to communicate, there is a possibility that the winning node may not have access to the client network, but won the arbitration process anyway. Because of the obvious limitation of this configuration, this method is not recommended.

Microsoft recommends configurations similar to configuration 1 and 2 above. Configuration 2 is more desirable with an extra private network in addition to the one mentioned in the example. This allows for two private networks for cluster communications and provides the ability to keep all cluster communications isolated. Private networks may be accomplished by the use of a crossover cable between nodes or by the use of isolated hubs.