Tip: Don’t locate the Windows 2000 paging (swap) file on a shared array. Place it on a local drive in each node in the cluster.
Explanation: Since the paging file is specific to a specific computer, not to the cluster, the paging file should be located on the nodes of the cluster, not the shared array. Not only will performance suffer if you do this, it could create cluster-related problems during failover.
Version: 7.0, 2000
Date Added: 5-23-2002
*****
Tip: The drive letter used for a shared array by SQL Server clustering must be the same drive letter on all nodes of the cluster
Explanation: When you configure your nodes for SQL Server clustering, one of your chores is to create a shared array for your instance of SQL Server, and then assign it a drive letter. It is very important that each node use the exact same drive letter for the shared array. If you don’t, then SQL Server clustering will not be able to fail over from one node to another.
Version: 7.0, 2000
Date Added: 7-1-2002
*****
Tip: If you want to initiate a manual fail over of SQL Server from one SQL Server clustered node to another, you must do this through Cluster Administrator, not Enterprise Manager.
Explanation: SQL Server clustering is designed in such a way as to only allow the manual failover of one node to another by using the Cluster Administrator. If you try to do this through Enterprise Manager by stopping the SQL Server services, manual failover will not occur automatically. What will happen is that the cluster service will automatically restart the service on the same node. If you try stopping the SQL Server services more than four times in a row, SQL Server may failover, but it also may produce some unexpected results. So don’t do it.
Version: 7.0, 2000
Date Added: 7-1-2002
*****
Tip: Under some circumstances, the private network connection of a cluster can switch from “Internal cluster communications only (private network)” to “All communications (mixed network)”. This can cause the network connection for the cluster to fail, preventing users from accessing your cluster.
Explanation: Windows 2000 uses a technology called Media Sense, which is part of the Plug and Play features of the operating system. The default behavior of Media Sense is to destroy a network card’s TCP/IP connection when the network connection fails (for whatever reason). When that happens, the Cluster Service receives a notification of the event and removes this network connection from the available connections. In most cases, this is not a problem.
Unfortunately, if network connectivity to both network adapters in a cluster node fail at the same time, the network role for the cluster can be changed from “Internal cluster communications only (private network)” to “All communications (mixed network).” When this happens, it is possible for the cluster resources to be bound to the network adapter designed for the private network, not the public network, which means users cannot access the cluster. In effect, the cluster has failed and no clients can connect to it.
In most cases, this problem can be fixed manually, by changing the private network connection role back to “Internal cluster communications only (private network),” and then rebooting all of the nodes of the cluster and bringing it back up properly. But this is a temporary fix. If this problem repeats itself, you have no choice but to turn of the Media Sense feature of the operating system. See this Microsoft Article on how to hack the registry to turn off Media Sense.
Version: 7.0, 2000 (Windows 2000 Advanced Server issue)
Date Added: 8-29-2002