VM/ESA 2.2.0 Performance Changes
- Performance Improvements
- Performance Considerations
- Performance Management
The VMCF modules have been changed to have full MP locking capability. As a result, VMCF functions no longer have to run on the master processor. This change does not have any appreciable effect on overall processor usage. However, it can result in improved throughput and response times for systems that are currently master processor constrained and that run applications that use VMCF (such as TCP/IP and OfficeVision/VM).
RDRLIST, FILELIST, PEEK, and their associated Xedit macros have been rewritten in REXX, performance-optimized, and compiled. This significantly reduces the amount of processor time used by these functions. Processor usage reductions ranging from 22% to 27% have been observed for RDRLIST and PEEK. The percentage reduction in processor usage for FILELIST is inversely proportional to the number of files being displayed. For example, a 17% reduction was observed for the case of 10 files, while a 9% reduction was observed for 1000 files.
The uncompiled source files are provided on the S-disk for customers who wish to make modifications. Customers with the REXX compiler are advised to recompile the updated files before placing them back into CMSINST so as to retain the performance advantages.
The elapsed time and processor time required to use the PEEK exec to view spool files created by the DISK DUMP command have been greatly reduced. The larger the file, the larger the percentage improvement. Percentage reductions exceeding 90% have been observed for disk dump files having more than 8000 records.
The implementation of the FILELIST option has been changed significantly, resulting in improved performance. Decreases in processor usage ranging from about 10% (few files) to about 40% (many files) have been observed.
The CMSINST logical shared segment has been moved above the 16 MB line, thus making more virtual storage available for data and applications that must reside below the line.
The following additional CMS system files have been moved into CMSINST:
ALIALIST EXEC PROFALIA XEDIT SAUTH XEDIT AUTHLIST EXEC PROFAUTH XEDIT SDIR XEDIT CSLLIST EXEC PROFCATL XEDIT SMODE XEDIT CSLMAP EXEC PROFCLST XEDIT VMLCAT XEDIT DIRLIST EXEC PROFCSLM XEDIT X$CLST$X XEDIT MACLIST EXEC PROFDLST XEDIT X$CSLM$X XEDIT NAMES EXEC PROFFDAT XEDIT X$DLST$X XEDIT NOTE EXEC PROFFSEA XEDIT X$MLST$X XEDIT SENDFILE EXEC PROFFSHR XEDIT X$NAME$X XEDIT VMLINK EXEC PROFMLST XEDIT X$NVML$X XEDIT X$CATL$X EXEC PROFNOTE XEDIT X$ONAM$X XEDIT X$NDIR$X EXEC PROFSEND XEDIT X$SEND$X XEDIT $@VML$NK XEDIT PROFVMLK XEDIT
When these functions are used, this change decreases per-user real storage requirements and eliminates the processor time and I/Os that were required to load them from the S-disk into memory.
The only sizable CMS areas that remain below the 16 MB line are 1 MB of the shared portion of the CMS saved system, the nonshared portion of the CMS saved system (72 KB), and file status table (FST) entries (including the HELP segment). FSTs for read-mostly shared minidisks can be removed from the user's primary address space by replacing these minidisks with one or more SFS dircontrol directories residing in VM data spaces.
CMS working set size and DASD page slot usage have decreased for the case where CMS multitasking functions have not been used during the CMS session. These improvements translate into reduced paging, lower processor usage, and improved response times.
The amount of improvement will typically be larger on systems that use expanded storage for paging. The removal of large numbers of one-time referenced pages has freed up much space in expanded storage that can now be used for other pages that are more likely to be re-referenced. This improved expanded storage efficiency allows for a higher percentage of total paging to be done in expanded storage, resulting in a sizable DASD page I/O reduction.
Most of the OS GETMAIN/FREEMAIN requests made by GCS has been eliminated through the use of preallocated storage, resulting in decreased processor usage. This change resulted in a 0.3% decrease in total system processor usage for the FS8F CMS-intensive workload.
CMS GUI performance has been improved in the following areas:
The performance of Xedit when invoked from the CMSDESK GUI application has been improved substantially. Average response time decreased by over 40% and host processor CPU usage decreased by more than 60% for a set of example Xedit interactions.
DtEventNotify is one of the functions provided by the distributed GUI toolkit (DT) API. The amount of host and workstation processing done by DtEventNotify has been reduced considerably (60% in one example) in the version of DT that is shipped with VM/ESA 2.2.0.
The DtEventNotify improvement, and most of the Xedit improvements, first appeared in the 6/96 Recommended Service Update to VM/ESA 2.1.0.
Some of the shared segments provided with VM/ESA have been moved to new locations. While installing VM/ESA 2.2.0., check to make sure that this has not caused any overlaps with other shared segments in your system.
A number of new monitor records and fields have been added. Some of the more significant changes are summarized below. For a complete list of changes, see the MONITOR LIST1403 file (on MAINT's 194 disk) for VM/ESA 2.2.0.
- Scheduler Data
The monitor previously recorded dispatch list priority, which is relative to ATOD (artificial time of day), but did not record the ATOD value. This made the priority information less useful. ATOD and ATOD2 (associated with limit list processing) have been added to records D0/R10, D2/R4, D2/R5, and D2/R6.
- Master Processor Data
The dispatcher uses a processor local dispatch vector (PLDV) construct to manage work from the dispatch list associated with a given processor. There is a PLDV for each processor. The master processor also has a second PLDV for master-only work. The monitor uses high frequency sampling to record the number of tasks on each of the PLDVs. For the master processor, this is the sum of both PLDVs. The number of tasks on the master-only PLDV is now recorded in D5/R3 in addition to the sum of both PLDVs.
There were no changes to monitor output as a result of year 2000 considerations because monitor uses 4-byte date fields.
None of the VM/ESA 2.2.0 performance changes are expected to have a significant effect on the values reported in the virtual machine resource usage accounting record. There have been no changes to the format of this record.
VM Performance Reporting Facility 1.2.1 (VMPRF) will run on VM/ESA 2.2.0 with the same support as VM/ESA 2.1.0. APAR VM60882 (PTF UM28172) is recommended. VMPRF at this service level includes the following enhancements:
- Support for >30000 userids.
- Four new reports.
- MINIDISK_CACHE_USAGE_BY_TIME (PRF103) supplements the information provided by MINIDISK_CACHE_BY_TIME, including information on MDC storage targets and usage for both main and expanded storage.
- SYSTEM_FACILITIES_BY_TIME (PRF104) contains CCW translation fast path performance information.
- SFS_BY_REQUESTS_BY_TIME (PRF105) contains a breakdown of SFS byte file system requests.
- USER_CONFIGURATION (PRF106) contains user virtual machine configuration and tuning settings by userid.
Realtime Monitor VM/ESA 1.5.2 (RTM/ESA) requires APAR GC05398 (PTF UG033816 to run on VM/ESA 2.2.0.
FCON/ESA Versions 2.3.00 and 2.3.01 will work with VM/ESA 2.2.0, but do not exploit any of the additional monitor data fields. With FCON/ESA Version 2.3.01, reports have been adapted to show the following new fields:
- Master-only PLDV statistics are included in PROCLOG display.
- Dispatch list priority and number of minor time slices used have been included in the 'Add to disp. list' and "Drop from disp. list' entries respectively of the UTRANDET display.
Performance Analysis Facility/VM 1.1.3 (VMPAF) will run on VM/ESA 2.2.0 with the same support as VM/ESA 2.1.0.