In this issue, I’ll cover the following:
1. JES2 R4 and R5 Caution
2. Cheryl Watson’s TUNING Letter, 1998, No. 1 Summary
3. Updates to Cheryl Watson’s TUNING Letter, 1998, No. 1
4. LE and PL/1 Overhead
5. New Seminars on Parallel Sysplex
1. JES2 R4 and R5 Caution
On page 13 of our 1998, No. 1 issue, we described some JES2 user experiences and advised caution using the JES2 R4 $ACTIVATE command. (It is needed before you can use some of the new R4 facilities, primarily WLM Batch Initiators.) Bob Shannon from Programart pointed out that conversations on the JES forums have indicated some possible problems with the $ACTIVATE for JES2 R4 and R5. WSC Flash 9817 describes the situation. (It is available to IBMers, but wasn’t on the Web as of noon on May 5. It should be soon). Essentially, IBM is recommending that you postpone a $ACTIVATE until you have applied the fixes for open APARs OW31341, OW29813, and OW31452. John Hutchinson of IBM’s JES2 team indicated that he expects the APARs to be closed by the end of the month. The problem, which may include corruption of the checkpoint data set, is not being seen in all sites, but if you haven’t done the $ACTIVATE, it might be better to wait and apply the fixes first.
2. Cheryl Watson’s TUNING Letter, 1998, No. 1 Summary
Cheryl Watson’s TUNING Letter, 1998, No. 1 was mailed to subscribers on 1 May, 1998 (we’ll be caught up really soon!). The following is an extract from the Management Summary to show you the scope of that 40 page issue. Cheryl’s List item 3 following this summary contains some updates to the 1998 No. 1 TUNING Letter issue.
One of the most significant items in this issue is a clarification of IBM’s January 13th announcement about Parallel Sysplex License Charge (PSLC) pricing. We’ve written about PSLC before in prior issues (Nov/Dec 1994, pages 12-15 and 1996, No. 3, pages 22-26). If you qualify for PSLC pricing you can obtain a 30% or more discount on all of your IBM software costs. The qualification rules for PSLC pricing were changed on January 13th, and there is a lot of misunderstanding about the announcement. Basically, IBM is now requiring sites that are getting multi-system PSLC discounts to prove that they are using a coupling facility in production. Some people seem to believe that this means they need to spend $100,000 or so to implement parallel sysplex. You really don’t need to do that. Our suggestions starting on page 7 show you how to qualify by running a small LPAR on one of the systems. You do not need to buy a stand-alone coupling facility in order to continue qualifying for the PSLC discount.
The focus of our last issue was Tuning Parallel Sysplex and it included some guidelines for the SYNC and ASYNC service times between a coupling facility and a host. A SHARE session by Kathy Walsh provided an update on those service times that can be used as the primary rule of thumb for tuning a coupling facility. These can be found on page 16.
CICS Recommendations
This issue contains several important items about CICS. The first item relates to large installations that want to move into goal mode and collect CICS V4 or CICS TS response times. These sites may experience some high overhead in CPU usage of the workload manager (WLM) that they might not be expecting. One site saw 40-50% of one CPU used by WLM. Our article on page 6 describes the situation, recommends a partial solution, and identifies the APAR to track for this condition.
If you’re interested in CICS and VSAM Record Level Sharing (RLS), you may be interested in our notes from a SHARE user experience on RLS starting on page 11. The section describes some important APARs to apply and track.
And finally, our series on CICS performance by Bob Archambeault provides some critical information on the use of CICS and Language Environment (LE). He had heard people complaining about the overhead in CICS when applications were converted to use LE, and he ran some tests to determine the difference. The differences were significant. As an example, VS COBOL II showed a 52% increase in CPU time when running in an LE/370 environment. The CPU time nearly doubled using COBOL for MVS (which uses LE) compared to COBOL II without LE. Bob also found that changing a simple programming option, CBLPSHPOP (PUSH and POP HANDLE), made a really dramatic difference in CPU times (from .7 seconds to less than 0.02 seconds). Please make sure that your CICS programmers get this article.
Configuration Changes and Parmlib
We’ve started a 2-part article on what items to change in parmlib and other areas when you make some type of configuration change: upgrading or downgrading a processor, adding or removing storage, upgrading a software release, changing an LPAR configuration, moving applications from one system to another, increasing or decreasing the volume of an application, adding or removing a system from a sysplex, or adding or removing users from a system. In many, many, sites we have found that the capacity was constrained because someone forgot to change a simple parameter after a configuration change. Most (but not all!) installations remember to increase options such as the domain constraints when a processor upgrade occurs, but did you know that there are over 25 parameters that should be reviewed after such an upgrade? If even one of them is forgotten, you may be limiting the work on your system. We think this will become one of the major checklists in your installation for any type of configuration change.
SHARE Trip Report
After each SHARE conference we provide selections from some of our favorite and most worthwhile sessions. As we’ve pointed out before, we consider SHARE to be one of the least expensive forms of education on a wide variety of topics. In this issue (starting on page 11) we provide extracts on the following topics: VSAM Record Level Sharing (RLS) user experience, DB2 for OS/390 R5 (including some important recommendations if you are considering data sharing), JES2 R4 user experiences, OS/390 recommended APARs and Web sites (including recommendations for improving GRS performance), several parallel sysplex items (new SYNC/ASYNC guidelines, dynamic CF dispatch enhancement, YEAR 2000 Sysplex Test Datesource, dynamic ICF expansion, and additional APARs), update on SmartBatch (a product to reduce the batch window), recommendation for use of discretionary in goal mode, pointers to Web pages for Internet capacity planning, and an explanation of the additions to RMF R4 (including some wonderful new measurements).
Elsewhere in This Issue
This issue is mailed with our new CPU Chart which contains updates for the newest Millennium series (and their surprising announcement that their 75 MIPS CMOS processor will be upgraded to 80 MIPS in April), a clarification of our opinions on IBM’s LSPR ratings of HDS and Amdahl processors, and a few minor updates. In our S/390 News, we have an important announcement about a change in our recommendation in keeping WLM type 99 records and our summary of the most recent Washington Systems Center and Storage and Network System Center flashes.
3. Updates to Cheryl Watson’s TUNING Letter, 1998, No. 1
The introduction and table of contents in TUNING Letter 1998, No. 1 indicated that an updated CPU Chart would be accompanying this issue. The chart has been delayed and will be mailed to subscribers by mid-May. If you need data on processors before that time, please contact us and we’ll be glad to help.
There were a few corrections that we’d like to point out on the 1998, No. 1 issue.
- The article on CICS, COBOL, and Language Environment (starting on page 37) made several references to LE/370. This was simply an example and the high CPU times found in LE/370 are also found in LE for MVS & VM. In fact, several of the tests were actually run on LE 1.6, which is LE for MVS & VM. The point is that any release of LE may result in the CPU overhead that Bob Archambeault described in this article.
- Two of the Web pages mentioned on page 18 have changed URLs since we went to press. To find Web Server sample programs, see the New Technologies Center site at http://www.s390.ibm.com/nc/sntc. The Web site indicated on page 18 for access to the ICSS (not ICCS) Redbook (SG24-2074) has changed. Instead, use the standard access to redbooks at http://www.redbooks.ibm.com/redbooksonline.html.
- An IBMer from the PSLC Software Pricing group looked at our article on page 7 and suggested some clarifications. We said that IBM was “changing” and “tightening” the requirements for qualifying for PSLC pricing. He pointed out that it’s really a clarification of a policy that had already been in place. He’s absolutely right! The reason that we included the article is that the policy was often not enforced by some local managers, and this “clarification” came as a big surprise to many S/390 customers. We also neglected to indicate that you must be running continually in production (We only mentioned being in production, not that it needed to be the normal mode of operation). Item 1 on page 7 should also indicate that the machines must be connected to a *common* coupling facility in order to qualify for this pricing discount.
4. LE and PL/1 Overhead
Here’s an item that is in the TUNING Letter that was just mailed. We think that Cheryl’s List subscribers who don’t get the TUNING Letter would benefit from it as well.
A customer called and described a problem when they migrated from SP 5.2.2 to OS/390 R3. (The same problem could occur during a conversion to R4 and R5 as well.) They saw a 25-35% increase in CPU time in several critical production PL/1 programs. The problem was in Language Environment (LE) 7.1 and 8.1. APAR PQ14007 describes the situation and mentions a job that went from 2.5 hours to 4.5 hours. The PTFs became available 98/04/09.
We’re currently working on an article on LE for an upcoming TUNING Letter issue since LE is turning out to be a serious performance problem. Many sites are migrating to COBOL for MVS using LE and finding a 20% increase in CPU time (some even see 30%). Tuning LE properly can reduce that overhead to closer to 5%, but it takes some work.
5. New Seminars on Parallel Sysplex
A new, free, Boole & Babbage seminar on Parallel Sysplex is coming to a site near you. “Whether you’re approaching the starting blocks or already on your way, this seminar will help you breeze to victory with your Parallel Sysplex implementation. Boole & Babbage has teamed up with both IBM and Watson & Walker consultant, Jerry King, to bring you up to speed on the very latest advancements in parallel computing. Through this educational seminar, you can gain the knowledge you need to help develop a winning Parallel Sysplex strategy — no matter where you are in the migration process. This no charge comprehensive half-day seminar guides you through emerging issues and solutions, including an overview of the Data Sharing technology introduced by IMS®, DB2® and VSAM. The Parallel Sysplex solution can make true 24×7 computing a reality. In just a few short hours, this intensive program can give you the insight to help you thrive in the leading edge S/390 environment.” For a schedule of seminars and more information, see Boole & Babbage’s Web page http://www.boole.com/events/Par_Sysplex/parsys_info.html. (Ed. Note – these seminars are no longer being given and the site is invalid. For upcoming seminars, you can check out <http://www.bmc.com>.)
That’s all for now. Stay tuned!