JD Williams, the big UK-based online retailer, was the first IT shop there to put the z196 into production, and central to its decision to buy was the availability of powerful specialty processors. As it turned out, the specialty processors, particularly the zAAP for running Java workloads, delivered sufficient MIPS to allow the company to order fewer general processor MIPS, effectively lowering the cost of the z196. See the full case study here.
The z196 specialty processors deliver 1,200 MIPS each, compared to 580 MIPS for the z9 specialty processors, which JD Williams had run previously. The same number of z196 zAAPs would deliver better than twice the number of specialty processor MIPS than before.
Data center managers assume that the z196 will be way too expensive. However, for organizations that analyze and plan their workloads with the goal of taking maximum advantage of the zIIP and zAAP specialty engines they can, as JD Williams discovered, boost their MIPS while getting a z196 for the cost of a z10.
The z196 was designed to tightly integrate with the full set of IBM System z assist processors, which are implemented in microcode on the new machine. JD Williams, from the outset, intended to take full advantage of the specialty processors to run its extensive Java workloads. Given the increased power of the specialty engines, JD Williams didn’t need to purchase as many general processor MIPS, which trimmed the acquisition cost considerably. And, as the company shifted more of the workload to specialty processor MIPS, it also could realize significant software licensing savings.
The initial JD Williams plan called for implementing as many 10 specialty processors. Since the z196 implements specialty processors through microcode, however, the company could start with just 4 zAAPs, 1 zIIP, 4 IFLs. If it wanted to add another specialty processor later, it could do so easily, without having to add more hardware or central processor MIPS first, as was the case with the z9. In the end, the company bought a 2-book, 28 processor z196 for the same cost as a z10 with comparable aggregate MIPS although they would be distributed differently between specialty and general processors.
The JD Williams experience shows the common assumption that the z196 will be outrageously expensive need not be true. You can juggle the mix of central and specialty processor MIPS to deliver the MIPS you need at the lowest possible price. Of course, you must plan workloads that make effective use of the specialty processors.
In the case of JD Williams those workloads were Java and WebSphere. For other organizations the workloads may be business intelligence, analytics, or Web 2.0. It pays to expand your thinking about what constitutes suitable mainframe workloads today. Between the lower cost specialty processor MIPS and the accompanying software licensing savings, a z196 running diverse workloads might be quite a bargain. And, if you consider running zPrime on the z196 specialty processors you can dramatically increase the amount of workloads you run and substantially boost your software licensing savings.
When the z196 BC models arrive later this year, lower cost models will be available . When the System z Solution Editions for the z196 arrive, the cost for new workloads will drop again.
Any organization can make the kind of deal JD Williams did. All it takes is planning your workloads to tap the specialty processors and shift cycles off the general processors. You’ll lower your acquisition cost and you’ll lower ongoing software licensing charges. If that’s not enough, throw in zPrime and amplify your savings even more. (In coming weeks DancingDinosaur will report on a zPrime user in Europe that is capturing huge savings in software licensing.)
