Managing Cluster Performance - Compaq ProLiant Clusters HA/F100 Administrator's Manual

5-12 Compaq ProLiant Clusters HA/F100 and HA/F200 Administrator Guide

Managing Cluster Performance

As applications or resources transfer from one server to another, the
performance of the clustered servers will likely change. The extent of the
change depends on how well equipped the other node is to handle the
increased workload. This is especially obvious after a server failure, where all
of the cluster resources may move to the other server.
Use the Windows NTS/E or Windows 2000 Advanced Server performance
monitor utility to observe and track system performance. Some applications
have their own internal performance measurement capabilities. Constant
monitoring of each system's performance in a cluster may identify a disparity
in the workload being performed by each server. Adjust the distribution of the
cluster applications so that the overall cluster performance is maximized.
Dynamic load balancing is not available in MSCS. However, MSCS can
statically load balance your cluster nodes. Information obtained from the
performance monitor utility can be used to determine whether either of the
cluster nodes is operating at too high a performance level. Then use Cluster
Administrator to fail over as many cluster groups as necessary to balance the
load of each cluster node. Additionally, disk I/O may be balanced across the
dual host bus adapters in a Compaq ProLiant Cluster HA/F200 configuration.
This functionality is only available when accessing two separate RAID arrays.
The timing of static load balancing depends on the type of group to be moved
and how many clients are using the group. File and print services are normally
not business-critical, nor do they sustain an extremely high utilization rate. For
these reasons, file and print services are good candidates to move whenever
static load balancing needs to occur, even when the overall cluster utilization
is high.
However, some applications, such as databases, should not be moved from one
server to another during peak processing periods. To move a database from
one server to another, the database must be shut down and restarted. Users do
not have access to the database during the time it takes the database to restart.
Move a database group to another server during non-peak hours.