Home 

Tuning Letter 

Classes 

Cheryl's List 
 
 

 
    Cheryl's List #7 - Oct. 2, 1997
 
     
 
    Welcome to Cheryl's List. If you obtained this from someone else and would like your own copy in the future (no charge!), see the information at the end of this document. Past issues can always be obtained at http://www.watsonwalker.com/archives.html. In order for any tables to appear correctly, please set your browser to show mail in a fixed width font. (In Netscape, that option is under "Options", "Mail and News Preferences", "Appearance".) 

    In this issue, I'll cover the following: 

    1.  Catalog CPU Overhead APARs 
    2.  Cheryl Watson's TUNING Letter, 1997, No. 3 Summary 
    3.  WSC Flash Update: COBOL and CMOS 
    4.  WE'RE HIRING! 
    5.  IBM Manuals on the Web 
    6.  MVS Expo 
    7.  Our Classes 
     

    1. Catalog CPU Overhead APARs 

    In the last two newsletters (and one yet to be mailed), we've discussed some catalog address space APARs. There's an update in each issue that we want to be sure you look at. In the TUNING Letter 1997, No. 2, page 5 we warned you about APAR OW26940 (97/07/02 - DFSMS 1.3). This APAR reports a 70% increase in CPU cycles in the CATALOG address space when moving to DFSMS 1.3. The fix for this APAR became available on June 17, 1997. 

    In the 1997 No. 3 issue (page 5), we also described OW27250 (97/06/17 - DFSMS 1.3), a similar type of APAR. Andrew Stewart, of Stelco Inc. in Canada, provided us some feedback on the effectiveness of the fixes to these two problems. He reported the following: 
     

    •  pre-OS/390 R3 (SP4.3) 
      • CAS CPU per day - 60 minutes 
    • June 1 - June 28 (OS/390 R2)
      • CAS CPU per day - 120 minutes 
    • June 29 applied OW26940 
      •  CAS - no change - 120 minutes 
    • July 13 applied OW27250 
      •  CAS CPU per day - 100 minutes 
       
    As you can see, the results of applying the fixes to these problems are not very encouraging. IBM is still working on the problem of why the CATALOG address space is consuming so much more CPU time. 

    In our 1997 No. 4 issue (will be mailed by next week), we report that, in some sites, these APARs result in significant increases (instead of expected decreases) in CPU cycles consumed by the CATALOG address space after migrating to DFSMS 1.3. One installation recently reported that it was OW26940 that caused an increase of 50% in CPU time consumed by CATALOG, and the remainder appeared to be due to OW27250. 

    IBM recently issued informational APAR II10752 to assist you in collecting the correct information for reporting this type of Catalog performance problem when running DFSMS 1.3. So if you are running into a problem like this we suggest you obtain a copy of this APAR and follow the instructions before reporting the problem to IBM. 
     
     
    2.  Cheryl Watson's TUNING Letter, 1997, No. 3 Summary 

    The 1997 No. 3 issue was mailed September 4. So that subscribers who haven't seen that issue yet (you know how slow internal routing can be) will know what's in it, and so non-subscribers can get an idea of the scope of our newsletter, we're including some of the highlights here taken from the Management Summary section. The 1997 No. 4 issue is expected to be mailed the second week of October. 

    UCB VSCR 
    Our Focus this issue is on UCB Virtual Storage Constraint Relief (VSCR) (also referred to as "UCBs above the line"). Over the years, IBM has introduced several options to reduce problems with UCBs, such as too many fixed frames below 16 MB and too few device addresses. There are solutions, but you'll need to put in some effort to implement these facilities. We're bringing UCB VSCR up now even though the facilities have been available for a couple of years, because few sites have taken the time to implement them. And almost every site we see needs help with UCB VSCR. We're sure the article on page 13 will help. 

    TCP/IP OpenEdition and Performance Improvements 
    We hear quite a few questions and concerns about TCP/IP performance on MVS. It doesn't have a good track record for performance yet and is of concern to most installations. IBM announced a new set of performance enhancements that may provide some relief. Some of the enhancements are available only to OS/390 R3 sites. See our description of these new features on page 20. 

    Alternate Master Catalog 
    "The catalog structure is an integral part of the operating system. If it breaks, the operating system dies." This is the introduction to our article on page 23 from Sylvia Gorman of NIU which describes how to create and maintain an alternate master catalog for emergency master catalog backup and recovery. If your installation does not currently have such a plan in place, this article could prove invaluable in reducing the downtime resulting from a broken master catalog. 

    CICS Performance APARs 
    Bob Archambeault, who has been writing the CICS series in the TUNING Letter, provides a variety of tips, suggestions, and performance APARs for CICS V4 and CICS Transaction Server (commonly referred to as Version 5). This is a very important article if you're running either of these CICS versions. 

    Ordering OS/390 R3 or R4 
    In the S/390 News on page 6, we discuss considerations for determining whether you should order OS/390 R3 or R4. Please look at this article if you are close to ordering either one of these releases. In summary, R4 has a higher release to release software license cost (ranging from 2.5% to 9%) but provides more performance benefits for OpenEdition and provides more Internet facilities. 

    S/390 News 
    In the last issue, we introduced some Catalog performance APARs. For some sites, they don't seem to be providing as much relief as we had hoped. In the S/390 news starting on page 5, you'll find an update and more APARs for catalog performance, a question for readers about WLM goal mode, some WLM and SRM APARs of interest, a new Web site for parallel sysplex, new OS/390 APARs, several OpenEdition performance APARs, a pointer to some excellent OpenEdition tuning documentation, more TCP/IP performance APARs, two ISPF performance APARs, a warning about the increased CPU time in ISPF V4 (from 5% to 12%), an informational APAR for DFSMS data compression, some IMS and DB2 performance APARs, and our summary of WSC Flashes. The WSC Flashes include an update to WSC Flash 9608 which describes the performance degradation on COBOL programs running on CMOS machines. We've discussed this in previous issues, and are happy to find that the flash indicates that the newest CMOS processors, the Rx5 series, appears to have made significant improvements in packed decimal processing. There is still a penalty for COBOL programs that use TRUNC(BIN) as a compile option for heavy subscript processing. 

    Parmlib Series 
    Our popular Parmlib series continues in this issue with a look at several of the parameters in IEAOPTxx. There are many parameters in this parmlib member and we will provide additional tuning recommendations for it in the next issue. You should find several recommendations in this issue to help begin tuning your OPT member now. 

    3. WSC Flash Update: COBOL and CMOS 

    Note: This was published in our 1997 No. 3 issue, but we wanted to include it here for continuity in the Cheryl's List discussion about COBOL and CMOS. 

    WSC Flash 9608 - Potential Performance Improvement for COBOL Programs on CMOS 

    This flash was updated 6/27/97 with some additional benchmarks using the ES/9672-RX5. The new timings show a significant improvement in handling of packed decimal instructions on a CMOS machine. If you remember from previous issues and Cheryl's List at http://www.watsonwalker.com/archives.html, a heavy use of packed decimal instructions took more CPU time on the CMOS processors from IBM and HDS than expected from IBM's LSPR numbers. IBM has been making headway in resolving this issue. 

    The latest Generation 4 processors seem to have greatly improved the processing time. The flash shows timings for various COBOL compile options with heavy subscript processing. From our evaluation, only the TRUNC=BIN feature still results in CPU times that are greater than expected. For example, IBM's LSPR ratings for the 972 and RX5 for CB84 (the COBOL workload) indicate that there is expected to be a 35% increase in CPU time for a single job. From the flash, a job that took 6 seconds on the 972 took 9 seconds on the RX5, but should have taken 8.1 seconds according to LSPR. (Note: IBM's rounding to whole seconds makes a fair analysis a bit difficult.) That amounts to an 11% overhead that wasn't expected. Realize that this is for a "worst case" program that does nothing but table processing. The other measurements show better results than expected (again, it's difficult to say because of the rounding). 
     

    4.  WE'RE HIRING!  

    We're looking for a SAS programmer with good MVS knowledge who wants to join us in our Sarasota, Florida office. 

    We need someone who is self-motivated and can do some or all of the following: 

    -  code, test, and help design new MVS SAS products (required) 
    -  support our current SAS products (required) 
    -  support OS/390 on our P390 
    -  help write documentation 
    -  conduct search 
    -  answer customer questions about performance and capacity planning 
    -  help teach and/or update sections of our OS/390 courses 

    C, assembler, or ISPF experience would be a bonus. Contact Tom Walker at tom@watsonwalker.com or fax a resume to: 941-366-6479. Please pass this request on to anyone you know who might be interested. 
     

    5.  IBM Manuals on the Web 

    IBM is investigating methods of providing documentation through the Web. We previously mentioned the availability of many manuals that are available from the IBM sites. You can obtain access to most OS/390 manuals via the Web by going to the OS/390 Home Page at http://www.s390.ibm.com/os390 and clicking on The Library icon. From there, you will find a list of groups of manuals that are available. While you are there, please take the time to give IBM feedback on whether you want manuals available in Adobe Acrobat PDF format. We have heard lots of complaints about the difficulty in printing softcopy manuals and we think that the PDF format may be a good solution. They need some feedback, though, if they are going to put the effort into the project. 

    IBM is also developing a new technique for using online manuals called a "Task Atlas". This is a technique where the manual creators encode the manuals with keywords and pointers to allow a selection of topics across all manuals. For example, to install a product, you often have to look at many manuals such as the installation guide, the SMF manual, JES manuals, a reference manual, and several others. Using the Task Atlas, you could simply indicate the product and "install" and you could have immediate access to just the sections of each manual that you need. In fact, you do not need to be aware that there are multiple manuals. The Task Atlas is currently available as a prototype from the OS/390 Web site (above). Try it and let them know what you think. 
     

    6. MVS Expo 

    IBM is holding their annual MVS Expo and Performance Technical Conference http://www.training.ibm.com/ibmedu/conf.htm/osmvs/ in Orlando, Florida this October 20-24. We'll have a booth at the technology fair, so please stop by and say hello. Peter Enrico and I will both be presenting sessions at the Expo: 

    Cheryl:  New OS/390 Key Performance and Capacity Indicators 
                  WLM Quickstart Policy Update and Recommendations 

    Peter:    Understanding OS/390 Coupling Facility Measurements 
                 Performance Considerations for Making OS/390 the Server of Choice in Your Shop 
     

    7.  Our Classes 

    All of our classes have been going well. We only have two more scheduled this year: "OS/390, Parallel Sysplex & Workload Manager" will be given in Sarasota October 27-31, and "Advanced OS/390 Performance & Capacity Planning" will be taught in Sarasota November 10-14. We've recently updated the course outline for the Advanced course based on student feedback, so be sure to check out the new outline on our Web page. 

    We've had several requests for the 1998 class schedule and we're still working on it. We're currently trying to obtain space for the classes from our local hotels. We have tentatively scheduled all of our classes for the May-June 1998 timeframe in Sarasota.  We expect to finalize the schedule for these classes by the end of October and will let you know via this listserver. 
     

    That's all for now. Stay tuned! 

    Cheryl Watson 
     

    ========================================================================== 

    Thanks for subscribing to Cheryl's List! Feel free to forward this to others who may be interested. 

    If you obtained this from someone else and would like your own copy in the future, just go to our web page http://www.watsonwalker.com and fill out the form under "Cheryl's List" -- that signs you up. Remember, it's a one-way list, from us to you. If you make a "reply", it will come just to me, not to the other members of the list. To subscribe or unsubscribe via email, send an email message with only the word "subscribe" or "unsubscribe" as the body of the message to <cheryls-list-request@xmission.com>. Past issues can always be obtained at http://www.watsonwalker.com/archives.html

    In these letters, we alert you to selected APARs, flashes, or manuals, answer questions of general interest, provide updates to our printed newsletter (Cheryl Watson's TUNING Letter), let you know if anything of importance has been added to our Web pages, and, of course, tell you about our other products and services. 

    Please note: This email service does not attempt to match the large scope and volume of information we provide in the TUNING Letter. (That publication comes out six times a year and costs $465, $515 abroad.) For subscribers, Cheryl's List will provide quicker corrections, time-sensitive updates, breaking news, or further comments on newsletter articles already in your possession. 

    For non-subscribers, these transmissions will give you a hint, we hope, of the quality and scope of the material normally found in the 40-60 pages of a typical TUNING Letter. If you don't subscribe, we'd love to have you as a customer. Please see our Web page for telephone numbers and the tables of contents of all past issues. By the way, ALL of the past print issues of the TUNING Letter are still available and can be purchased by anyone. 

    So - we hope you'll find this service valuable. Be sure to send email to cheryl@watsonwalker.com if you have any questions or comments.    To the Top