|
|
| |
 Creating a Service Definition
| | |
 |
|
I don't want to print out my service definition using SCRIPT. How can I convert this to an HTML file?
(July, 2000)
Create a SCRIPT/VS print file, just as you always did in the past (by using the "Print as GML" option on the "File" pulldown menu in the WLM ISPF application). Use the B2H tool to convert this file to HTML.
The B2H tool is downloadable as a
.zip package from the VM Download Library. See the
Description of B2H for more information.
|
 | |
Why is it so important to keep the number of service classes to a minimum? Is this to reduce WLM overhead?
(Q3 from the OS/390 Expo and Performance Conference, October 1998, Atlanta)
No. There are a couple of reasons why it helps to have no more than 25 to 35 active service class periods at any point in time:
- WLM responsiveness
At each 10 second policy interval, WLM selects the service class period in the most need of help and takes one policy action on its behalf. The more active service class periods you have that need assistance, the more 10-second cycles it will take WLM to take all the needed policy actions for these service classes. This could reduce WLM's responsiveness in managing to goals.
- Better WLM decisions
As part of its decision-making, WLM makes projections for how work will behave when resource changes are made. The accuracy of these projections depends on the quality of the history data gathered for a service class. The more like work that can be put into a single service class, the more statistically valid will be its history data, and the better will be WLM's decisions.
| |
| |
How do I make sure my "loved ones" get the best possible treatment in goal mode? I worry that I will lose control of them under goal mode. Should I put them in SYSSTC?
(Q5 from the OS/390 Expo and Performance Conference, October 1998, Atlanta)
Firstly, make your "loved ones" the most important work in the system, typically Importance 1 or 2. This means that the "loved ones" goals will be honored over all other goals. Secondly, set velocity or response time goals that reflect your true requirements for this work. Thirdly, opt for response time goals if you can, especially for CICS or IMS. When CICS or IMS regions are recognized as transaction servers with response time goals, WLM gives them protective storage targets even before they encounter page delay problems.
SYSSTC will certainly give the "loved ones" extremely favorable CPU access. However, if you are running storage-constrained with CICS or IMS, you may be better off with aggressive response time goals. This way you retain the proactive storage protection afforded to servers with response time goals.
If you want no limits to the level of service for the "loved ones", you could set very aggressive or even unattainable goals. This further reduces the possibility that the "loved ones" would ever become a resource donor to other work in the system. Keep in mind, though, that the "loved ones" would then be allowed to consume your system when the workload increases.
| |
| |
What workloads are currently exploiting enclaves? How do I know if they are being used on my system?
(Q8 from the OS/390 Expo and Performance Conference, October 1998, Atlanta)
As of OS/390 V2R6, the workloads listed below use enclaves. Enclaves represent transactions that do not have address space boundaries. They can span many address spaces or execute concurrently in a single address space. WLM is able to manage and account for these enclave transactions separately from the address space where they are executing. You can expect the list of exploiters to grow as new types of workloads are ported to the S390 platform:
| Workload |
WLM subsystem type |
| DB2 distributed queries |
DDF |
| DB2 sysplex query parallelism |
DB2 |
| Lan Server for MVS |
LSFM |
IBM HTTP Server for OS/390
(formerly Lotus Domino Go Webserver)
|
IWEB |
| WebSphere Application Server for z/OS |
CB |
| MQSeries Workflow |
MQ |
If you have work running in the service classes associated exclusively with any of the WLM subsystem types listed above, you have enclaves! In goal mode, if you do not classify work that is using enclaves, it defaults to the SYSOTHER service class with a discretionary goal. This service class can be used as an indicator of unclassified work. Keep it at zero activity!
If you have long-running enclaves, you can display the currently active ones using RMF Monitor III Delay report. This report displays the generic category *ENCLAVE. Using cursor-sensitive control, you can get a list of individual enclaves and, using cursor-sensitive control again, get a pop-up window showing classification details on the enclave.
| |
| |
How does WLM manage initiators for Unix System Services?
(Q9 from the OS/390 Expo and Performance Conference, October 1998, Atlanta)
As of OS/390 V2R4, WLM rather than APPC creates address spaces on demand for Unix System Services fork and spawn processes. WLM does not dynamically manage the number of these spaces as it does for batch initiator management. They are created strictly on demand.
When a forked child process ends, its address space is returned to a pool to be used by a new forked child. If these address spaces go unused for 30 minutes, they are terminated. If you want to limit the number of address spaces created for forked children, use the MAXPROCSYS parameter in the BPXPARMxx parmlib member. See OS/390 OpenEdition Planning for the details (SC28-1890). This book also has good information on how to classify UNIX System Services work.
These address spaces, sometimes called OE initiators, are classified by WLM under the OMVS subsystem type. The transaction name (that is, jobname) is by default the userid plus a number 0-9. However, if an enclave is active at the time of the fork or spawn, the child process runs under the enclave's service class and managed to the enclave's goal. For example, if an Webserver transaction makes a CGI request, the resulting child process is managed and reported as part of the original Webserver transaction and does not use the initiator's service class. Therefore, you would find service consumed by the CGI processing reported in RMF under the service class of the Webserver transactions.
| |
| |
I set my response time goals based on the observed average response time from RMF, but my online work eats up my system during busy times, and no other work gets done.
Set response time goals based on the response time during peak periods only. Using a long-term average response time that includes periods of high and low usage gives you a goal that is too aggressive for the peak periods. Select the response time that was achieved during the heaviest peak period to ensure you have a goal that is achievable. A similar method can be used with velocity goals, that is, select the velocity that was achieved during the heaviest peak period. This yields the lowest velocity that still meets the needs of the workload.
|
|
| |
Can you mix goal types for service class periods that have the same importance?
Yes. Service class periods are compared one to another by calculating a performance index (PI) for each period. The PI gives WLM a common way to track how well the work is doing regardless of goal type. The PI is 1 if the work in the period is meeting its goal exactly. The PI is less than 1 if the work is doing better than its goal. The PI is more than 1 if the work is missing its goal. For work at the same importance level, WLM attempts to equalize the PIs.
|
|
| |
What are the advantages of an enterprise-wide workload management policy?
If you have an enterprise-wide service definition, where service class goals, importances, and classification rules are the same across all sysplexes, you can move workloads among systems more easily. The relative business importance of the work is preserved across the move.
|
|
| |
How can I manage the individual transactions within the HSM address space?
Because individual HSM transactions are not known to WLM, HSM can be managed only as an address space with a single velocity goal. It is classified under the STC subsystem type.
|
|
|
|
| |
|
|
|
