Changes to IBM VM memory management for mainframes' larger workloads

To better support memory-intensive and varied virtual machines, IBM has improved the memory management technologies in its z/VM mainframe operating system.

One enhancement in the latest version of z/VM, 6.3, is that it does a better job balancing memory pages than z/VM 6.2, said Bill Bitner, senior software engineer, IBM Systems & Technology Group, while presenting at the SHARE IT 2013 conference in Boston.

Here are 10 changes in z/VM 6.3, released in July 2013, that affect virtual machine (VM) memory management:

1. Large IBM VM workloads performed better on the 6.3 version of z/VM compared to older versions of the OS, according to company tests. In z/VM versions 5.4 and 6.2, large workloads would become sluggish and unresponsive. IBM's z/VM 6.3 scales better; the largest partition can have up to 1 TB of memory.
2. With older versions of z/VM on mainframes, administrators have to provide double the paging space needed for virtual storage on a direct access storage device (DASD). With z/VM 6.3, IT no longer needs to double the paging space on DASD, but will need to pay special attention to the paging system configuration.
3. In z/VM 6.3, there is one global aging list (GAL) for the entire mainframe. This enables large memory support, and takes over the memory page filtering process that expanded storage (XSTORE) previously performed. Mainframe administrators can tune the global aging list's settings, but IBM strongly recommends keeping the defaults.
4. You should configure all expanded storage as central storage (CSTORE) in 6.3. While z/VM 6.3 supports XSTORE, IBM will drop it in future versions. The GAL helps approximate which pages were least recently used, rendering the expanded storage method of sorting memory pages unnecessary.
5. You can turn off early writes to DASD, but don't forgo DASD completely in environments that have ample memory. Even if you are not using memory over commit to share memory among VMs, you should write to the DASD.
6. VM needs can differ greatly -- consider the difference between an Oracle database VM that needs 40 GB of memory and an HTTP Web server that requires only 1 GB. Because it is difficult to overcommit memory well with such varied VM loads, z/VM 6.3 writes out the memory before you need to reclaim it.
7. To save space, storage slots on DASD no longer have to be contiguous in z/VM 6.3. If all spool and page space is used, z/VM will ab end. With non-contiguous slots, ab ends are less likely to occur.
8. The latest version of z/VM also uses real units for resource measurements, as opposed to calculations of frames. Information is reported in bytes, not pages. "Down the road, pages may come in different 'sizes and colors'," said Bitner.
9. There is a new option for IBM VM Hiper Dispatch workload management technology. With topology awareness, HiperDispatch will place VMs in accordance with their processor usage.
10. Private virtual disks (VDisks) join the available frame types, along with user frame lists and shared VDisks.