HP e-Commerce / XML director server appliance sa8250 - Users Guide

C H A P T E R 2 Load Balancing Across Multiple Servers
41
Load
Balancing
Across
Multiple
Servers
Balancing Algorithms
The SA8250 provides a choice of load balancing algorithms.
Services can be separately configured to load balance using a round-
robin or a response time algorithm. In most networks, the best
performance results from use of the response time algorithm. Under
this algorithm, the SA8250 measures the response time of each
request to each server in the server farm. It then balances requests to
the service among the servers, sending more requests to the fastest
servers and fewer to the slower ones, thus optimizing the average
response time.
In cases where OPR is used in unpredictable WAN environments,
response time metrics may be obscured by WAN latency variance. In
these situations, round-robin load balancing can provide equal
distribution of client requests to each fulfillment server.
The balancing algorithm is specified with this command:
config policygroup <name> service <name>
balancing [robin | load]
Response-time Metrics
For both balancing algorithms, servers can be assigned target
response times. These values indicate the desired average response
time for requests for specified services to be fulfilled, and instructs
the SA8250 to use alternate resources for fulfillment if the average
response time exceeds target response time. Target response time is
controlled with this command:
config policygroup <name> service <name>
response <mil-seconds>
If the servers do not meet the specified response time threshold,
backup servers, if available and enabled, are activated. In addition,
the servers providing lower priority services are throttled if the
response time is still not being met (if
throttle is enabled in the
policygroup). Both mechanisms are available for both of the load-
balancing algorithms.