Performance Improvements
The following items improve performance:
- Enhanced Timer Management
- Improved Utilization of Large Real Storage
- Improved Linux Guest QDIO Performance
- Large Volume CMS Minidisks
- TCP/IP Stack Improvements
Enhanced Timer Management
With z/VM 4.3.0, the CP timer management routines no longer use the master processor for multiprocessor serialization but instead use the scheduler lock. This eliminates the master processor as a potential system bottleneck for workloads that generate large numbers of timer requests. An example of such a workload would be large numbers of low-usage Linux guests. See Enhanced Timer Management for further discussion and measurement results.
Improved Utilization of Large Real Storage
CP has been changed in z/VM 4.3.0 to more effectively use real storage above the 2G line in environments where there is little or no expanded storage. In prior releases, modified stolen pages were paged out to expanded storage (if available) or DASD. With z/VM 4.3.0, when expanded storage is unavailable, full, or nearly full and there are frames not being used in storage above the 2G line, page steal copies pages from stolen frames below 2G to these unused frames above 2G. See Improved Utilization of Large Real Storage for measurement results.
Improved Linux Guest QDIO Performance
A combination of enhancements to z/VM, the Linux QDIO driver, and the OSA-Express microcode can increase the performance of Linux guests using Gigabit Ethernet or Fast Ethernet OSA-Express QDIO. Throughput increases up to 40% and CPU usage reductions of up to one half have been observed with these changes. There are no appreciable performance benefits until all 3 changes are in effect. Refer to OSA Express QDIO Performance Enhanced in Consolidated Linux under z/VM Environments for further information.
The VM updates are integrated into z/VM 4.3.0. They are also available via APAR on prior releases (VM62938 for z/VM 4.2.0; VM63036 for all other releases back to VM/ESA 2.4.0).
Large Volume CMS Minidisks
The performance of CMS minidisk file commit has been improved. Formerly, file commit caused all file allocation map blocks to be written out to DASD. With z/VM 4.3.0, just those allocation map blocks that have been modified are written back. This change has little effect on the performance of most minidisks (1000 cylinders or less) but can significantly improve the performance of file commit processing for very large minidisks. See Large Volume CMS Minidisks for measurement results.
TCP/IP Stack Improvements
CPU usage of the TCP/IP VM stack virtual machine has been reduced significantly, especially for the case where it serves as a router. CPU time reductions ranging from 24% to 66% have been observed. See TCP/IP Stack Performance Improvements for measurement results.