Health Check; Persistency - Avaya P330 User Manual

Health Check

Persistency

Avaya P330 Load Balancing Manager User Guide
The load balancer constantly checks the RSs to ensure that each RS is
accessible and operational. An RS that fails the health check is
automatically removed from the load balancer's internal list of currently
available RSs, and traffic is redirected to other available RSs.
There are several types of health check methods that the load balancer
can use, including:
• ICMP Ping - Each RS is periodically pinged. If no answer is
received, the RS is not operational.
• TCP Port Checking - A TCP connection is periodically opened
to each RS, checking for successful completion of the connection.
For FWLB, checking the firewalls is insufficient. The health checks must
be performed on the entities beyond the firewalls as well. In order to
ensure that the health check packets traverse the same firewall in both
directions, the packet's source and destination IP addresses should be the
IP addresses of the load balancer interfaces on each side of the firewall.
For each load balancer, both the local and remote addresses must be
configured. In addition, the load balancers on both sides of the firewall
must be configured symmetrically.
For non-transparent FWLB (with NAT), there is only one load balancer.
In this case, you must configure an IP address beyond the firewall as the
health check address. Like other non-transparent FWLB sessions, the
health check session returns through the same firewall according to the
NAT address it was given.
Persistency is the maintenance of the connection between the server and
the client over multiple sessions. Persistency ensures that all traffic from
the client is directed to the same RS.
Persistency is achieved by using naturally persistent load balancing
metrics (such as Hash or MinMiss hash) or by forcing persistent load
balancing decisions on non-persistent load balancing metrics (such as
Round Robin). Persistency is forced by storing the history of the latest
decisions in a cache for a limited time, and then sending the packets to
the appropriate RS according to the previous load balancing decisions.
16