Using HP Global Workload Manager with Serviceguard
7
Shared Resource Domains
The recommended approach in most situations for configuring gWLM in a HA Serviceguard 
environment is to define a separate shared resource domain for each unit of hardware (complex or 
server) in the cluster. A shared resource domain is a collection of partitions that share resources. 
gWLM manages the resources within a shared resource domain.
A complex can be subdivided using nPartitions or virtual partitions into multiple hosts each of which 
can participate as SG cluster nodes. As resource allocation is performed independently for each 
shared resource domain, aligning each shared resource domain with independent hardware provides 
maximal availability and avoids the issue of attempting to communicate and share resources with a 
node that may have failed. You can deploy shared resource domains on one or more of the nodes in 
the cluster depending upon your needs:
• A shared resource domain on the primary node enables dynamic resource allocation between the 
active applications.
• A shared resource domain on the fail-over server of an active/active cluster configuration enables 
those low-priority jobs on the fail-over node to better utilize the resources when they would 
otherwise go unused.
• A shared resource domain on the fail-over system in an active/stand-by configuration can enable 
the dynamic activation of TiCAP resources and thereby minimize the cost of the fail-over system 
except during times of an active failover.
It is not necessary for the Central Management Server (CMS) to be a member of the Serviceguard 
Cluster that it manages. While the CMS must be up and running in order to make changes to the 
gWLM configuration, it does not require the same level of HA as the gWLM agent. The other 
on-going function of the CMS is the collection and storage of historical performance data. If, 
however, the CMS is unavailable for a period of time, the gWLM agent daemons queue up the 
performance data for later transmission to the CMS.
Workload and Package Concepts
In most circumstances you will treat each of your Serviceguard packages as a workload. This enables 
you to associate a gWLM policy with each package to determine the amount of resource it should 
receive. While not absolutely required, it provides the most direct and logical way in which to think 
about resource allocation to the packages. Because the package can run on different nodes with 
potentially different resource availability, this is actually realized in the gWLM configuration as a set 
of workloads—one for each compartment (such as a virtual partition or nPartition) in which the 
package may be started in each shared resource domain. 










