Description

Explains what a Predictor is in relation to Load Balancing and VIPs.

Content / Solution:

A Predictor is the load-balancing method or algorithm that determines how the loadbalancer selects a real server in a server farm to service a client request.  We currently offer two methods:

  • The Least Connections predictor instructs the loadbalancer to select the server with the fewest number of connections.
  • The Round Robin predictor instructs the loadbalancer to select the next real server in the list of real servers. Each "hit" to the VIP will result in the next available server in the pool.

However, if the Predictor is being used in conjunction with a Persistence Profile, note that any persistence characteristcs will take precedence over the predictor algorithm. If any criteria in a session matches the persistence profile, additional hits to the VIP will continue to be directed to the same Real Server until the persistence criteria is no longer met.This can result in scenarios where traffic from multiple different sessions stays connected to a given Real Server.

For example, if you have persistence set to an IP method and you're browsing from a typical home/office network where a single IP address is being shared among different systems behind the router/firewall, then each connection will look as if it came from the same IP address. In such a scenario, each connection (regardless of how many systems or browsers are tried) will automatically bound to the same real server in the pool to which the predictor directed the original request and will continue to be directed there until the persistence criteria are no longer met.