Using HP Global Workload Manager with Serviceguard

5
• Simplifying routine/scheduled maintenance—With gWLM running on a secondary server, you can
fail over packages from their primary servers to perform maintenance on the primary servers,
knowing gWLM will ensure that the applications are allocated CPU resources based on their
relative priorities. Thus, your most critical applications will be getting the most CPU resources. After
all the work is finished, you can move the applications back to their primary servers.
Figure 2 illustrates some of these benefits. Upon a failure on one of the nodes, the package is started
on one of the other systems in the cluster. gWLM controls the level of resources available to the
packages, based on their business priorities and the load on each application. In this example Pkg A
is moved to Node 2 due to a failure on Node 1.
Figure 2. Failover with adjustment to resource allocation
The next scenario, depicted in Figure 3, shows a 12-node Serviceguard cluster built from seven
systems. This configuration contains six Serviceguard primary servers labeled App1-6, each hosting a
single application, and six Serviceguard secondary servers hosted by six virtual partitions (vPar 0-5)
that handle failovers from any of the six primary servers. In addition to hosting the secondary servers,
the system hosts vPar 6 for running low-priority, noncritical jobs and is not a cluster member. All the
secondary servers remain in idle standby during normal operations, awaiting a package failover.
gWLM is configured and running on each vPar continuously optimizing the allocation of the resources
of the system and enabling vPar 6 to utilize any excess resources that would otherwise go unused.
When App 3 fails over to vPar 2, gWLM detects the resource demand on the server and allocates
additional cores from the other vPars. If multiple packages become active on the secondary servers,
gWLM distributes the available resources among the servers in use, shifting whole cores among vPars
to meet the configured policies. When App 3 returns to its primary server, gWLM can again reduce
the CPU resource allocation to vPar 2, reflecting its reduced need for CPU resources. As a result,
gWLM allows for a much less expensive HA solution: The failover environment for many active servers
can share resources to minimize costs—and even allow the resources to be used to by noncritical jobs