Introduction to the z/VM large memory tests
The objective of the z/VM large memory - Linux on System z project was to analyze the results observed with Linux® guests running a database server in a z/VM® environment using a relatively large amount of main memory (80 GB) and then also overcommitting that memory. We compiled an executive overview of our z/VM large memory performance test run results.
Objectives
- VM Resource Manager Cooperative Memory Management (VMRM-CMM), initially available with z/VM 5.2 (also known as CMM Version 1)
- Collaborative Memory Management Assist (CMMA), initially available with z/VM 5.3 (also known as CMM Version 2)
These features were expected to improve the overall system performance in cases where the overall z/VM system is constrained for real storage.
This paper shows the performance of the Linux database guests under an online transaction processing (OLTP) workload using asynchronous disk I/O. The number of guests was scaled to create the memory overcommitment on z/VM. The starting point for the scaling was a setup with multiple Linux guests that allocated all available memory and then the number of guests was increased until the memory overcommitment of a factor of two was reached.
Executive summary
Both z/VM 5.2 and z/VM 5.3 were able to handle the scenario with a relatively large amount of physical memory (80 GB) and the use of twice as much virtual memory than physically available (memory overcommitment). We recommend the usage of z/VM 5.3 and VMRM-CMM for this type of workload (transactional database workload using asynchronous disk I/O). In the memory overcommitted case, z/VM 5.3 shows significant improvements for performance compared to z/VM 5.2. The features VMRM-CMM and CMMA require that the most current fixes available for these features for Linux and z/VM are installed. Memory overcommitment is expected to cause a performance degradation, where VMRM-CMM reduces that impact to 15%, which is a very small impact for a memory overcommitment of a factor of two.