You can configure vertical stacking to enable the application placement
controller to start more than one instance of the dynamic cluster to improve
bottleneck conditions within an application.
Before you begin
Vertical stacking is supported on the z/OS® operating
system, but the WebSphere® Application
Server for z/OS multiple
servant feature is the preferred alternative. Consider using vertical stacking
for application servers on the z/OS operating systems only for application
servers that host applications that are constrained to run in a single servant
application server.
About this task
Use vertical stacking to improve bottleneck conditions in your application.
With vertical stacking, multiple server instances can start on a node. By
configuring multiple server instances, you can use all the power that is available
on the nodes when a large workload exists for the application.
Procedure
- Determine whether your application has an internal bottleneck by
profiling the application. The application placement controller
assumes that the only resource bottleneck that an application might have is
either in the processor or in memory. Otherwise, the application might have
an internal bottleneck.
- Run an instance of the application on each node.
- Increase the load for the application.
- Using the runtime reports, the throughput curve and processor
utilization on a node that is running the application. To
view the runtime topology, click Runtime operations > Runtime topology.
If the throughput curve saturates while increasing the load, but
the processor utilization remains well below 100%, the application might have
an internal bottleneck. Configure vertical stacking to improve these bottleneck
conditions. If the server instances are able to reach processor utilization
near 100%, do not enable vertical stacking.
- Determine the appropriate stacking number. The stacking number
is the number of application servers for a cluster that are needed to use
all the power of a node. For more information about determining
a stacking number, see Determining the stacking number for a dynamic cluster
.
- Enable vertical stacking and the stacking number. If
your dynamic cluster is deployed to a node group that has homogeneous nodes,
you can configure the stacking number one time in the administrative console.
If your dynamic cluster is deployed to a node group that has heterogeneous
nodes, you must configure the stacking number by specifying custom properties.
Results
When a large workload exists, the application placement controller
can fully use the power of the nodes by starting multiple cluster instances
on each node.
The general formula to determine the percentage of the processor
that is used by each instance on a node is 100%/maximum_number_of_stacked_instances.
For example, if three instances are stacked, then workload is throttled to
prevent any one of these instances from using more than 33% of the processor.
What to do next
Tune and monitor your application placement configuration.