Contents | Previous | Next

Summary of Key Findings

This section summarizes z/VM 5.2.0 performance with links that take you to more detailed information.

z/VM 5.2.0 includes a number of performance-related changes -- performance improvements, performance considerations, and changes that affect VM performance management. The Enhanced Large Real Storage Exploitation support affects all three of these in significant ways. CP was changed so that even though most of its code continues to run with 31-bit addressing, it is now able to work with guest pages without first having to move them to frames that are below the 2 GB real storage line. Furthermore, the CP code and most of its data structures can now reside in real storage above the 2G line. As a result, workloads that show a 2G-line constraint on prior releases should no longer have this constraint on z/VM 5.2.0. Furthermore, measurement results demonstrate that z/VM can now fully utilize real storage sizes up to the supported maximum of 128 GB.

The extensive CP changes required for this storage constraint relief necessitated some unavoidable increases in CPU usage. Our regression results reflect this. The workloads experienced increases in total CPU time per transaction ranging from 2% to 11% with the 3G high-paging workload at the top of this range. Specialized workloads that target a narrow range of CP services can show differences outside this range, including a net performance improvement when the CP services being used have benefitted from one or more of the z/VM 5.2.0 performance improvements.

z/VM 5.2.0 includes support for QDIO Enhanced Buffer State Management (QEBSM), a hardware assist that moves the processing associated with typical QDIO data transfers from CP to the processor millicode. Measurement results show reductions in total CPU usage ranging from 13% to 36%, resulting in throughput improvements ranging from 0% to 50% for the measured QDIO, HiperSockets, and FCP workloads.

With APAR VM63855, z/VM 5.2.0 now supports the use of Parallel Access Volumes (PAV) for user minidisks. Measurement results show that the use of PAV can greatly improve the performance of DASD volumes that experience frequent I/O requests from multiple users.

The performance report updates for z/VM 5.2.0 also include four performance evaluations that can be helpful when making system configuration decisions. Linux Disk I/O Alternatives shows relative performance for a wide range of methods that a Linux guest can choose to do disk I/O. Dedicated OSA vs Virtual Switch Comparison, compares two methods of providing network connectivity to Linux guests. Layer3 and Layer2 Comparisons, compares the Layer3 and Layer2 OSA-Express2 transport modes for the following cases: 1) z/VM virtual switch, 2) Linux directly attached to an OSA-Express2 Gigabit Ethernet card (1 Gb and 10 Gb). Finally, Guest Cryptographic Enhancements, discusses the performance of the additional cryptographic support provided in z/VM 5.2.0, including support for the Cryptographic Express 2 coprocessor (CEX2C) and the Cryptographic Express2 Accelerator coprocessor (CEX2A).

Contents | Previous | Next