The current implementation of the backup server requires that all non-backup servers fail prior to directing requests to backup servers. This may not allow for maintaining the same level of performance in the server farm. The ability to maintain same performance level for a given service is critical for many customers.
Per Server Based Real Server Backup allows the backup servers to be associated with the specified primary servers. When a primary server fails, its backup server starts processing the traffic no matter what state the other primary servers are in. This feature works with the current real server back mechanism, by providing additional control of the backup server selection.
Currently, when a primary server goes down another server is selected among the active primary servers. Until all the primary servers are down, the server is selected from the backup servers. Additionally, the users can configure
backup-stay-active to keep the server selection in the backup groups active, even when some primary servers come back up.
With this feature, the associated primary and backup servers back up each other, regardless of the state of the other service ports. If a backup server is associated with a primary server, they work as a pair so each can substitute the other when it becomes unavailable.
If the backup-stay-active is configured, the backup server continues to process the traffic even after the primary server comes up again.
If the server selection predictor is least connection, the backup server may be overwhelmed by the flood of the new connections when its primary server goes down. The same is true when the primary server goes back up and starts to take over the connections from the backup server. The slow start mechanism will be used whenever the switching of the backup or primary server happens, to give the server the chance to ramp up.
