|
Contents | Previous | Next
Performance Improvements
The following items improve performance:
- Storage Management Improvements
- Collaborative Memory Management Assist
- Improved MP Locking
- Diagnose X'9C' Support
- Improved SCSI Disk Performance
- VM Guest LAN QDIO Simulation Improvement
- Virtual Switch Link Aggregation
z/VM 5.3 includes several important enhancements to CP storage
management: Page Management Blocks (PGMBKs) can now reside above the
real storage 2G line, contiguous frame management has been further
improved, and fast available list searching has been implemented.
These improvements resulted in improved performance in
storage-constrained environments, greatly increased the amount of
in-use virtual storage that z/VM can support, and allowed the maximum
real storage size supported by z/VM to be increased from 128 GB to 256
GB. See Improved Real Storage Scalability
for further discussion and performance results.
Collaborative Memory Management Assist
This new assist allows virtual machines to exploit the new Extract
and Set Storage Attributes (ESSA) instruction to exchange information
between the z/VM control program and the guest regarding the state and use of
guest pages. This function requires z/VM 5.3, the appropriate hardware,
and a Linux kernel that contains support for the Collaborative Memory
Management Assist (CMMA). A performance evaluation was conducted to
assess the relative merits of CMMA and VM Resource Manager Cooperative
Memory Management (VMRM-CMM), another method for enhancing memory
management of z/VM systems with Linux guests that first became
available with z/VM 5.2. Performance improvements were observed when
VMRM-CMM, CMMA, or the combination of VMRM-CMM and CMMA were enabled on
the system. At lower memory over-commitment ratios, all three
algorithms provided similar benefits. For the workload and
configuration chosen for this study, CMMA provided the most benefit at
higher memory over-commitment ratios. See Memory
Management: VMRM-CMM and CMMA for further discussion and
performance results.
A new locking protocol has been implemented that reduces contention
for the scheduler lock. In many cases where formerly the scheduler
lock had to be held in exclusive mode, this is now replaced by holding
the scheduler lock in share mode and holding the new Processor Local
Dispatch Vector (PLDV) lock (one per processor) in exclusive mode. This
reduces the amount of time the scheduler lock must be held exclusive,
resulting in more efficient usage of large n-way configurations. See Improved Processor Scalability for further
discussion and performance results.
Diagnose X'9C' is a new protocol for guest operating systems to
notify CP about spin lock situations. It is similar to diagnose X'44'
but allows specification of a target virtual processor. Diagnose X'9C'
provided a 2% to 12% throughput improvement over diagnose X'44' for
various measured Linux guest configurations having processor
contention. No benefit is expected in configurations without processor
contention. Diagnose X'9C' support is also available in z/VM 5.2 via
PTF UM31642. Linux and z/OS have both been updated to use Diagnose
X'9C'. See Diagnose X'9C' Support for further
discussion and performance results.
z/VM 5.3 contains several performance improvements for I/O to
emulated FBA on SCSI volumes.
- z/VM now exploits the SCSI write-same function of the IBM
2105 and 2107 DASD subsystems, so as to accelerate the CMS FORMAT
function for minidisks on SCSI volumes.
- CP modules that support SCSI were tuned to reduce path length for
common kinds of I/O requests.
- For CP paging to SCSI volumes, the paging subsystem was changed to
bypass FBA emulation and instead call the SCSI modules directly.
These changes resulted in substantial performance improvements for
applicable workloads. See SCSI Performance
Improvements for further discussion and performance results.
The CPU time required to implement VM Guest LAN QDIO simulation has
been reduced. We observed a 4.6% CPU usage decrease for an example
workload that uses this connectivity intensively. In addition, the no-contention 64 GB Apache run shown in
the Improved Real Storage Scalability discussion has improved
performance in z/VM 5.3 due to this improvement.
Link aggregation allows you to combine multiple physical
OSA-Express2 ports into a single logical link for increased bandwidth
and for nondisruptive failover in the event that a port becomes
unavailable. Having the ability to add additional cards can result in
increased throughput, particularly when the OSA card is being fully
utilized. Measurement results show throughput increases ranging from
6% to 15% for a low-utilization OSA card and throughput increases from
84% to 100% for a high-utilization OSA card, as well as reductions in
CPU time ranging from 0% to 22%. See Virtual
Switch Link Aggregation for further discussion and performance
results.
Contents | Previous | Next
|
