CLUSTERS


Choosing the server availability threshold
Setting the server availability threshold on each server is a key factor in balancing the workload in the cluster. Setting the server availability threshold too high can result in failover occurring unnecessarily. Setting it too low can result in poor performance for users who could have received better performance from a different server.

To determine the proper value for the server availability threshold, do the following:

1. During periods of normal to heavy load, use one of the following methods to observe the server availability index:

2. Set an initial availability threshold based on the results of your observation. Consider the following when setting this value: 3. Track other cluster statistics to see if the workload is reasonably balanced. The following table lists some of the statistics that are helpful in determining if the workload is balanced.
Statistic nameDescription
Server.AvailabilityIndexThe current value of the server availability index. The values range from 0 to 100. A value of 0 indicates that there are no resources available on the server. A value of 100 means that the server is completely available.
Server.ExpansionFactorThe current expansion factor. This value is used to compute the availability index. A value of 1 indicates that the server is completing transactions at the minimum time for that server. A value of 64 indicates that it is taking 64 times longer than the minimum time to complete transactions. By default, an expansion factor of 64 indicates a fully-loaded server, and results in an availability index of 0 (zero). You can use the NOTES.INI setting Server_Transinfo_Range to change the expansion factor value that indicates a fully loaded server.

You can use the Show AI server command to obtain a suggested availability index.

Server.Cluster.OpenRedirects.LoadBalance.SuccessfulThe number of times a BUSY server successfully redirects a client to another cluster member.
Server.Cluster.OpenRedirects.LoadBalance.UnsuccessfulThe number of times a BUSY server is unsuccessful in redirecting a client to another cluster member. A server will be unsuccessful if the target server is also in a BUSY state or otherwise unavailable.
Server.Cluster.OpenRequest.ClusterBusyThe number of times a BUSY server tries to redirect a client request when all other cluster servers were also BUSY.
Server.Cluster.OpenRequest.LoadBalancedThe number of times a user tried to open a database on this server when the server was BUSY.

4. Compare these statistics for all the servers in the cluster.

5. Adjust the server availability threshold on any servers that do not seem well balanced.

Note Workload balancing is not a substitute for having adequate computer resources for your enterprise. If your servers are already overloaded, workload balancing merely increases the problem because there is no place for a BUSY server to send client requests. Looking for an available server, however, decreases the performance on an already busy server.

Related topics