In this issue, I’ll cover the following:
1. CMOS Update
2. WSC Flashes of Interest
3. Updates to our Web site
4. IBM-MAIN Listserver address
5. IBM Manuals Online
6. Class Brochures now available
7. Cheryl Watson’s TUNING Letter, 1996 No. 6 status
1. CMOS Update
My item on “CMOS and COBOL” in the last Cheryl’s List caused quite a stir! Amdahl had included some of my comments in an “internal-use-only” Newsflash, but in a context which implied that I thought that IBM’s Rx4 CMOS machines caused a 200% increase in a user’s program. My original comment was referring to pre-Rx4 machines, not Rx4. Amdahl has retracted the statement, but there’s still a bit of confusion out there. If you have any questions about what I meant, please contact me.
With that said, however, let me point out two other items relating to CMOS. First, I don’t think I’ve ever stressed enough the fact that the biggest culprit in most of the overhead problems is the misuse of the COBOL compiler option called TRUNC(BIN). This option causes additional decimal arithmetic to be invoked when using subscripts, and it’s the decimal arithmetic instructions that are slower on the CMOS machines. For more efficient methods, use an installation default of TRUNC(OPT) or TRUNC(STD). See the COBOL manuals for additional information. Since some overhead for TRUNC(BIN) is present even in bipolar machines, you should ensure that TRUNC(BIN) is NOT the installation default.
The second point I’d like to make is that I’ve had very good reports on the Rx4 CMOS machines. In every case but one, I’ve heard that the Rx4 is meeting expectations based on LSPR ratings. In the one case, all programs but one are behaving as expected. The one program is taking 14.5 hours of CPU time instead of the expected 12 hours (a 20% increase). It turns out the program processes 85 volumes of tapes and does a pack and unpack on all of the tape data in order to perform compression. While we aren’t certain that this is the same phenomenon as the COBOL subscripts, we assume it is. In this installation, the program was scheduled to be re-written, and their schedule has simply been moved up.
2. WSC Flashes of Interest
I found two WSC Flashes issued in December to be especially interesting. Flash #9648 describes some problems that occurred when TCP/IP V3R2 was installed on the OS/390 R2 ServerPac. You need the instructions from the flash in order to complete the installation of TCP/IP. I’ve talked to at least four people in the middle of their OS/390 R2 install who were unaware of the flash.
The second flash, #9646, provides documentation about IBM’s CMOS SAP (System Assist Processor). The SAP acts as an offload engine for the CPUs, such as performing the functions of an IOP (I/O Processor), internal disk controller for the Multiprise 2000 series, and support for non-synchronous data sharing in a parallel sysplex environment.
A little-needed option of the 9672 CMOS machines allows you to configure one of your normal CPU processors to an additional SAP for I/O-intensive workloads, such as in a TPF environment. The flash explains the difference in the Configuration Specification setting of “I/O Intensive” or “Standard (processor intensive)”. In summary, the flash recommends that you use “I/O Intensive” for TPF environments, for Multiprise 2000 servers with internal disks, for MVS running under VM (this last one is for the pre-Rx4 machines only), or for any other extremely high I/O concentrated usage; use “Standard” for all of the rest. Note that “I/O Intensive” converts a regular CPU into a SAP, thus reducing the processing capacity of the machine. When I talked to the author of the flash, WSC’s Walt Caprice, he said that this seems to be a confusing area for most people and he wrote this flash to help people understand it.
3. UPDATES TO OUR WEB SITE
During this coming week, we’ll be uploading new course descriptions containing slight modifications to the topics and schedules. As mentioned in the last “Cheryl’s List”, we also added our Quickstart Policy paper and tables to our Web site.
4. IBM-MAIN LISTSERVER ADDRESS
In some issues of our Cheryl’s List #1, item 4 had a incorrect reference to “listerver@ua1vm.ua.edu.” It should have read “listserv@ua1vm.ua.edu.”
5. IBM Manuals Online
IBM is providing a Web page to allow you to access OS/390 manuals online. See the OS/390 home page athttp://www.s390.ibm.com/os390 and click on the new “library” icon.
6. Class Brochures Now Available
Our class brochures for 1997 will be printed in the next week. They’ll be sent to TUNING Letter subscribers with the next issue (see the next item), to members of “Cheryl’s List”, and to anyone in our contact database (prior students or anyone who has called for information). If you don’t get one in the next couple of weeks, please call. You can also obtain the information from our Web pages.
7. Cheryl Watson’s TUNING Letter, 1996 No. 6 status
The 1996 No. 6 issue is at the printer and we expect to mail it next week, with our new 1997 class brochures. So that subscribers will know what’s coming and non-subscribers can get an idea of the scope of our newsletter, we’re including a summary of the contents here. It’s taken from the “Management Issues” section that’s in each TUNING Letter.
Focus: RAID
“In late April of 1996, IBM withdrew the 3390 models 1, 2, and 3 from general availability, effectively ending a thirty year lineage of native count-key-data (CKD) mainframe storage devices which began with 2311s during the System/360 era. In their place, all of the mainframe DASD vendors have introduced storage subsystems based on fixed block architecture (FBA) devices which emulate CKD commands as well as virtual 3380 and 3390 device form factors. While these new subsystems offer substantial price performance and environmental benefits, they present a new class of performance issues which must be analyzed through a measurement system (i.e., RMF device activity monitoring) which was designed to report on the physical CKD paradigm.”
This is the introduction by well-known I/O authority Dr. H. Pat Artis to one of his articles on RAID in this issue. These new DASD subsystems provide reliability and stability by use of redundant systems (RAID – Redundant Array of Inexpensive (or Independent) Devices). When you add to that the fact that these new subsystems provide reductions in floorspace, energy requirements, and price/GB you’ll see why they’ll be in most installations within the next couple of years.
RAID may require changes in your current cache tuning techniques. If you’re trying to tune your RAID systems using traditional tuning methods, you could be doing exactly the opposite of what’s recommended for RAID. Because this is such an important subject, we’ve asked Dr. Artis to provide us with an introduction to RAID technology (RAID in a Nutshell), an article on how RAID is changing the analysis of our DASD subsystems (Sibling Pend – Like a Wheel Within a Wheel), and answers to some of the most frequently asked RAID questions from our users.
TCP/IP
TCP/IP was the Internet’s and UNIX’s choice of network protocol (similar to the IBM mainframe’s world of SNA). As companies are connecting more and more distributed systems to their mainframe, TCP/IP is becoming the protocol of choice in most installations. Because TCP/IP was designed for environments other than S/390, the performance under MVS or OS/390 has left a great deal to be desired.
IBM’s latest version of TCP/IP V3R2 is a major step forward in improving that performance. If you are beginning to have a large TCP/IP workload, you should be looking into migrating to the new release (we also include a quick product review). Also, InterLink Process Systems produces an MVS and OS/390 version of TCP/IP, TCPaccess, that often provides better performance than IBM’s version. Their newest version is also described here. In responding to their customers’ request for help in tuning TCP/IP, IBM has provided a TCP/IP Performance Tuning Guide and a Web site devoted to TCP/IP issues. If you haven’t encountered TCP/IP yet, plan on it soon. Every installation I talk to today is trying to learn how to manage this new, growing workload.
S/390 News
We’ve merged the “News of MVS” section and the “Ongoing Topics” into a single section called “S/390 News.” In that section, we mention that IBM’s WLM home page (http://www.s390.ibm.com/products/wlm) now contains a new article on WLM measurements and a pointer to our Quickstart Policy (first described in the May/June 1995 issue and the basis for the majority of goal mode service policies today). We also include in S/390 News the following: some useful WLM tips and APARs, an HCD SPE that improves HCD reporting, a pointer to HCD newsletters, a reminder about an old WSC flash that’s still applicable for CMOS capacity planning, forty net locations (including one on how to locate an IBM employee’s email id), an update on Year 2000 from IBM, pointers to some Year2000 conferences, a pointer to a Web page containing ESO pricing, and an announcement of a new quarterly IBM publication on networking-related stories about the Internet and Intranets (online or free subscription).
I’d like once more to stress the importance of planning the Year 2000 migration. The Year 2000 conferences, IBM’s essential white paper on Year 2000, and many vendors are now available with tools and consulting to help you move into your migration. If you haven’t begun your planning for one of the largest conversions you’ll ever go through, it’s important that you start now. In general, this means moving to at least OS/390 R2, the latest subsystem releases (CICS, IMS, DB2, etc.), and the latest compilers (such as COBOL for MVS – described in this issue with a list of possible migration tools). I don’t mean to belabor this point, but I’m still seeing less than 20% of my students are from sites that have a Year 2000 migration in progress.
More Tips
In our “What’s New” section, I describe the newest Sysplex Timer, the new TCP/IP releases, and the RMF cache reporter. For product reviews in this issue, we highlight three products: 1) PAI/O Driver from Performance Associates that provides benchmarks on DASD subsystems and provides assistance with defining performance guarantees in contracts; 2) PerfMan from The Information Systems Manager that provides a cost-effective method of managing Windows NT servers; and 3) IAM from Innovation Data Processing that can reduce response times, space usage, and CPU usage for programs accessing high activity VSAM files.
In our Q & As, we discuss how to tune TCP/IP, how to set the SRM parameter RCCCPUT, give a description of the CPU and storage overhead when moving to new releases of MVS (e.g. you could expect to see a CPU increase of 2% to 5% and a storage increase of 6MB to 10MB when moving from SP 4.3 to SP 5.2.2 or OS/390 R1), and provide some help with COBOL migration tools. And you’ll find Part 3 of Bob Archambeault’s series on CICS VSAM file tuning.
That’s all for now. Stay in touch and stay tuned!