February, 2010
It's the first meeting of the year, and yes, we were lucky enough to entice Tom Halinski to give up snow shoveling duty in Michigan to present to us AND to offer Compuware as our host. Compuware provided a wonderful facility and a great continental breakfast and fajita lunch.
Tom started off the morning with "Performance Improvements from the "Things I Wish They'd Told Me 8 Years Ago" - Visualized. The goal was to look at performance tips and do the tests himself to prove that these tips were good ones. He began by giving us a levelset, explaining the way DB2 SQL calls worked and then got right to the proofs. He showed that in fact, all the tips provided by Bonnie Baker, IBM, did prove themselves out, which should help DB2 performance people feel more assurance when they implement them. In fact, stage 2 predicates are much more CPU-intensive than stage 1, clustered indices actually make a major difference in execution time over non-clustered, and async I/O can be 94% faster than sync.
Dan Hoffman, Neon Corporation, presented "zWhole Wide World of Specialty Processors." Dan has been product manager at Neon for 3 years; prior to that, he was with BMC Software for 18 years. He presented the history and technical detail on all the specialty processors offered by IBM, beginning with the ICF. A key point is that all these engines are identical to GP engines; they are disabled for I/O interrupts, but not I/O. But the price is key. He explained that IBM offered these only for new workloads where there was competition for server selection; this gave mainframes an important edge. He then went into explaining how pricing for software works, leading into a description of the Neon zPRIME product and what kinds of cost improvements can be netted by IMS, CICS, DB2, batch and TSO/ISPF workloads. He noted that though the documentation specifies that you should not enable zPRIME for other vendors' software, there is no other bar to this option.
Lunch arrived early, so we adjourned briefly to fill our plates, then had a short but insightful presentation by Compuware. Our equally short business meeting covered the new discounted CMG membership program, a call for speakers and volunteers for CMG 2010 and a call for writers to contribute to Measure IT. There was also a solicitation for new board members for the region; after 16 years, Cathy would like to see someone else have the opportunity to lead NCCMG.
Doug Berg, Software Engineering of America, brought his great energy and good humor to that coveted "after lunch" spot, with "Best Practices in Batch Tuning - Empowering IT Staff to Reduce CPU and Run-Time in the Enterprise." He described himself as "the new kid on the block," but his experience kept showing - new to SEA perhaps, but not new to mainframes. He noted that since more than 50% of critical work is done by batch, which is growing while the window shrinks, it is important to ensure that tuning is done. Just because it isn't broken, that doesn't mean there aren't opportunities to make it really hum. But the challenge is that no one really has the time, unless they have automation and a great set of tools. Doug defined what you need to look for, indicating that you can hope to achieve 10-15% reduction in CPU use and elapsed time, which can translate, in some companies, to millions in savings in just a matter of months. Some of the areas he talked to were JCL efficiencies, exploiting data in memory, optimizing I/O, unnecessary data passes and data creation, slow utilities, failed jobs, waiting. He recommended creation of a knowledge base to guide tuning and automating whatever processes you could automate.
Jozo Dujmovic, San Francisco State University, brought us a very different presentation - the results of his research, "Benchmarking the Efficiency of Array Processing for Various Types of Language Processors." Jozo is a frequent speaker at National CMG and has contributed a lot to the field through his research of areas that validate the work we all do day to day. In this presentation, he showed how one actually does a proper benchmark and explained the various types of benchmarks and the pros and cons of each. In this case, he compared compiler performance across different languages for processing arrays. The body of the presentation was showing how he constructed the benchmark and how arrays work. The results were interesting and clearly showed that the choices one makes as to languages and compilers can make a huge difference in performance and resource demand for the resulting programs.
Finally, we were privileged to welcome the 2009 Mullen award winner, Chris Lynn, Safeway, presenting his award-winning paper, "Load Testing is Easy; Good Load Testing is Not. Preparation is the Difference." He defined 9 steps needed to plan for a successful experience, beginning with the obvious, but frequently omitted step - understanding what the goal of this testing should be. By taking the time upfront to plan, each testing effort is much more productive and the resulting data is more useful. The large number of questions showed the relevance of this topic to our audience.
Our next meeting will be May 4, 2010 - location to be determined. Any questions or comments, please contact Cathy Nolan at (cathy.nolan@comcast.net or nccmg@cmg.org).
Denise P. Kalm
Secretary, NCCMG