Memory requirements for z/OS in a multi-servant environment
User is looking for guidance about JVM heap size requirement when using multi-servants configuration in z/OS.
User's question is whether a 8GB max heap is really required in multi-servants configuration.
Steps to Reproduce
The JVM HeapSize is one component of sizing WebSphere z/OS Multi-Servant sizing.
Assuming a JVM MIN HeapSize = JVM MAX HeapSize then this amount of memory would be the “building block” as it relates to dedicated resources for each Servant. The number of Servants would be a multiplier using this “building block”.
With 64-bit environments, the minimum recommended JVM HeapSize is 4gb for small workloads and/or small memory footprints for each task.
Depending on the workload fluctuations and the memory footprints of each task running would impact the optimal JVM HeapSize.
In the field today, with Pega7, customers are opting to go with large JVM HeapSizes to reduce the number of JVMs required to manage.
For WebSphere z/OS there is another component available to consider besides JVM HeapSize. There is also the MIN-MAX Servants.
- HeapSize allocations for each Servant should be backed by real storage and dedicated CPU / Memory resources for the sizing of the servant.
- HeapSizes also have a MIN HeapSize / MAX HeapSize. For PRPC on z/OS, Pega recommends MIN HeapSize=MAX HeapSize. However, if MIN HeapSize is less than MAX HeapSize this introduces a third scaling component.
- Min/Max Servants allows WebSphere to add / remove servants as workload demands. The advantage of this better management / utilization of allocated resources (CPU / Memory).
With both of these sizing options, the client can adjust the JVM HeapSize and/or Min-Max Servants based on the workload demands comparing Peak to non-Peak.
- If the Min-Max Servants are equal then increasing the JVM HeapSize will reduce the number of address spaces and remain constant (i.e., constant amount of dedicated resources).
- If the Min-Max Servants are not equal reducing the JVM HeapSize will increase the number of address spaces as workload increases and reduce the number of address spaces as workload decreases (i.e., fluctuating amount of dedicated resources).
There are trade-off considerations.
- As JVM HeapSize increases, there is a potential of a higher density of workload on each Servant which may be impacted in the case of a Servant outage.
- As JVM HeapSize decreases, there are more address spaces to manage. And, each address space increases the amount of additional real and virtual resources.
Published March 29, 2018 - Updated October 8, 2020
0% found this useful
Have a question? Get answers now.
Visit the Collaboration Center to ask questions, engage in discussions, share ideas, and help others.