Coyote Point Systems Equalizer GX Series, Administration Manual

Page: 432 / 866

Server Pools and Server Instances
Configuring Server Pool Load-Balancing Options
Configure load balancing policy and response settings for each server pool independently. Multiple
clusters do not need to use the same load balancing configuration even if the same physical
server machines host them. For example, if one cluster on port 80 handles HTML traffic and one
on port 8000 serves images, you can configure different load balancing policies for each server
pool.
When you use adaptive load balancing (that is, you have not set the cluster’s load balancing policy
to round robin or static weight), you can adjust Equalizer to optimize performance.
Equalizer’s Load Balancing Policies
Equalizer supports the following load balancing policies, each of which is associated with a
particular algorithm that Equalizer uses to determine how to distribute requests among the
servers in the server pool:
l
Round-robin load balancing - distributes requests equally on the server pool in the cluster.
Equalizer dispatches the first incoming request to the first server, the second to the second
server, and so on. When Equalizer reaches the last server, it repeats the cycle. If a server
in the cluster is down, Equalizer does not send requests to that server. This is the default
method.
The round robin method does not support Equalizer’s adaptive load balancing feature; so,
Equalizer ignores the servers’ initial weights and does not attempt to dynamically adjust
server weights based on server performance.
l
Static load balancing - distributes requests among the servers depending on their assigned
initial weights. A server with a higher initial weight gets a higher percentage of the incoming
requests. Think of this method as a
weighted round robin implementation. Static weight
load balancing does not support Equalizer’s adaptive load balancing feature; Equalizer does
not dynamically adjust server weights based on server performance.
o
Adaptive
load balancing - distributes the load according to the following per-
formance indicators for each server.
o
Server response time
is the length of time for the server to begin sending reply
packets after Equalizer sends a request.
o
Active connection count
shows the number of connections currently active on the
server.
o
Server agent value
is the value returned by the server agent daemon (if any) run-
ning on the server.
l
Response
load balancing - dispatches the highest percentage of requests to the server with
the shortest response time. Equalizer does this carefully: if Equalizer sends too many
requests to a server, the result can be an overloaded server with slower response time. The
fastest response policy optimizes the cluster-wide response time. The fastest response
policy also checks the number of active connections and server agent values (if configured);
but both of these have less of an influence than they do under the adaptive load balancing
432 Copyright © 2014 Coyote Point Systems, A Subsidiary of Fortinet, Inc.