Use memcachedNodes parameter only or set primary node using both memcachedNodes and failoverNodes?
If we have 10 memcached nodes running on 10 machines and have 50 tomcat nodes runing on another set of 50 machines (Sticky session is used). There are following two ways to configure memcachedNodes and failoverNodes. 1. Configure each tomcat node to have only the memcachedNodes parameter with the 10 memcached nodes as value. 2. Configure each tomcat node to have both memcachedNodes parameter and failoverNodes parameter. The memcachedNodes parameter use the 10 memcached nodes as value and the failoverNodes use 9 of the 10 memcached nodes as value. In other words, specify one primary node for each tomcat node. Which option is better from performance point of view? Further to this, if the cluster starts with 2 memcached nodes and 10 tomcat nodes, automatically scales out to 10 memcached nodes and 50 tomcat nodes at peak time, and automatically scales in to 2 memcachend nodes and 10 tomcat nodes when the load drops to minimum. Which of the above option is better for implementing the auto scale in and scale out capability described above in terms of performance? Which one is better in terms of easy to implement?