CLUSTERS
Distributing mail databases
Clustering mail databases provides high availability to users. Some companies set up a cluster for mail only. This is not required but is one way to arrange your organization. Because mail is an important application, it is a good idea to have your mail databases in a cluster, whether or not the cluster is dedicated to mail servers only.
When you create a mail cluster, distribute the replicas for each server among the other servers in the cluster. This ensures that all the other servers share the load when a server fails, thus balancing the workload and keeping performance as high as possible.
It is a good idea to distribute replicas evenly across servers, as long as the servers are fairly equal in resources. If your mail cluster contains four servers and 1,200 users, for example, place 300 mail databases on each server, and place 100 replicas from each server onto each of the other three servers. When a server fails, 100 mail users fail over to each of the other three servers, increasing each server's work load by 33%. You might be tempted to place all 300 replicas from Server 1 onto Server 2 and all the replicas for Server 3 onto Server 4. In such a case, however, if Server 1 fails, all 300 users fail over to Server 2, increasing the workload on Server 2 by 100% but not increasing the workload on Server 3 or Server 4 at all.
The following figure shows a mail cluster that contains four servers with 300 mail databases on each server. Replicas of the mail databases are evenly distributed among all the other servers in the cluster, keeping the workload of the other servers as low as possible, even when failover occurs.
The following figure shows a mail cluster that contains two servers with 100 mail files on each server. Because there are only two servers, each server must fail over to the other server. Therefore, each server contains replicas of all the mail databases on the other server.
Since users often open mail databases once a day and leave them open, distributing the mail databases is usually adequate for workload balancing. You do not usually have to use separate workload balancing settings, especially if you dedicate servers to mail only.
After failing over to a replica mail database, users automatically return to the mail database on their mail server the next time they start their IBM® Lotus® Notes® clients, as long as the Location document that points to that mail database is the current Location document.
Note If you do not create a dedicated mail cluster, you should distribute mail databases equally among the cluster servers, if the cluster servers are approximately equal in power. If some servers are more powerful than others, distribute more databases to the more powerful servers. This distribution helps to keep the workload balanced.
Caution If you plan to create a cluster that includes servers running different releases of IBM® Lotus® Domino®, keep the following in mind:
When clustering applications, some applications need a higher level of availability than others. In addition, some databases are used much more frequently than others, and some require more computing resources to run than others. You should consider all of these factors when distributing application databases. With application databases, workload balancing is more important than with mail databases.
As with mail, you should try to distribute workload evenly across the cluster. However, take into consideration the power of each server. If some servers are more powerful than others, be sure to increase their workload appropriately. Also, take into consideration the amount of usage for each database.
The following figure shows a cluster with four servers of varying amounts of power. The databases in the cluster are distributed in a way that takes advantage of the resources of each server.
The following figure shows a cluster with four servers that are equal in power. The databases in this example all receive a similar amount of use. DB1 is a critical database, so each server contains a replica.