News

Analysis: System z10 Drastically Alters the Mainframe Market

• By Stephen Swoyer
• March 25, 2008

The new System z10 mainframes announced last month provide a striking example of this phenomenon. Big Blue announced five different z10 models, which, depending on configuration, offer more than 100 different capacity options -- including a behemoth-like 64-way setting.

The upshot, industry watchers say, is that -- similar to its z9 Enterprise Class (EC) and Business Class (BC) mainframes (which boasted unprecedented levels of capacity configurability) -- IBM has again drastically altered the rules of the mainframe game.

"The z10 will make the mainframe a much more attractive platform, especially for hosting new and emerging workloads, based on Java and/or Linux," says Dave Jones, a consultant and developer with V/Soft, a zLinux and z/VM-oriented mainframe ISV.

Jones managed to get up close and personal with a System z10 mainframe at last month's SHARE event. He came away impressed.

"IBM's claim of one System z10 [being equal to approximately]1500 Intel x86 servers is conservative," Jones told Enterprise Strategies. "Not only is the z10's processor more technically advanced -- for example, [there's] hardware support for decimal floating point instructions -- and faster [the z10 clocks in at 4.4-GHz, compared to the z9-EC's 1.7 GHz.], there are now four cores on a single chip, with a significantly improved 'super-scaler' instruction pipeline."

Gartner analysts -- and veteran mainframe watchers -- John Phelps and Mike Chuba did some number crunching of their own. Their conclusion: the System z10 is nothing like the z9 EC.

"The z10 EC offers more than 170 percent of the maximum computing capacity of the z9 EC -- to more than 25,000 millions of instructions per second (MIPS) -- due to an estimated 50 percent boost in processor performance and added engines," they point out, noting that the z9 EC supports a maximum of 54 processors.

One upshot of Big Blue's Big Iron performance overhaul is that the brawniest of System z10 configurations could prove attractive for some nontraditional applications -- such as high-performance or technical computing.

"The underlying technology has taken a major step up in chip frequency to 4.4 GHz in the z10 EC, which could provide additional performance for CPU-intensive workloads that historically have not been the mainframe's sweet spot," the analysts note. "The entry-level System z10 starts at approximately 200 MIPS, with 1.5TB of memory plus a new separate 16GB Hardware System Area standard in the base configuration. The z10 EC extends the platform's strengths in power, cooling and space efficiency, and delivers price/performance improvements in maintenance and software pricing."

Industry veteran Charles King is also impressed.

"What are we to make of [System z10]? In essence, … IBM has substantially increased the density of Big Iron while also making it more agile and better able to support new business processes," he argues. For this reason, King says, the high-end System z10 EC doesn't just change the rules of the mainframe game: it alters the data center status quo, too.

"In the System z10 EC and its attendant applications, tools, and services, IBM believes that organizations will find the foundation of New Enterprise Datacenter and answers to a host of critical business and technology challenges," he continues. The big question, according to King, is whether non-mainframe shops will take notice; the faithful will certainly be excited about -- and probably anxious to deploy -- System z10, but what about non-mainframe customers?

"[Existing mainframe customers] understand the mainframe's notable benefits and are likely planning the purchase and deployment of new z10 EC systems," he says. "Preaching the mainframe's value to non-users tends to be hard practice on the best of days. For IBM, the most practical and challenging course is to please its supporters, engage the curious, inform the unknowledgeable, and actively prove naysayers wrong. The z10 EC offers an array of new features aimed at improving performance and lowering overhead:"

Boosters point to an array of other z10 enhancements, such as: more than 50 new instructions (which collectively improve compiled-code efficiency); support for a 1 MB page frame size (which reduces memory management overhead); the addition of InfiniBand coupling links (which help accelerate parallel sysplex intersystem communications); HiperDispatch (which Big Blue says reduces task switching overhead); and a hardware decimal floating point implementation that IBM claims provides as much as a tenfold improvement over its predecessor in instruction execution.

What's a big mainframe shop to do? That's easy, Big Iron-watchers say: seriously consider System z10.

"Organizations with thousands of installed mainframe MIPS [should] examine the z10 EC for the capacity it offers, weighing carefully the potential hardware, software, and maintenance pricing benefits," write Gartner's Phelps and Chuba. For smaller organizations, of course, the z10 -- even in less brawny Business Class configurations -- might not be as smart an option as the eminently configurable z9 BC.

"The z9 BC may still represent the lowest cost and greatest granularity, but recognize that IBM could fill out the low end of the z10 EC some time during 2009 with a z9 BC replacement," they indicate. "Steer clear of commitments to the z9 beyond three years, particularly if capacity growth is anticipated, as the z9 is likely to be withdrawn from marketing during 2009 or 2010."

There's another wrinkle here, Phelps and Chuba say: the z10 builds on Big Iron's integrated power and cooling proposition. For this reason, it might also behoove organizations sensitive to such issues -- i.e., rising power and cooling costs -- to investigate System z10.

"Organizations sensitive to power, cooling and electricity limitations [should] evaluate further investment in the IBM mainframe as a vehicle to further consolidate workloads and reduce energy costs, and possibly avoid costly data center expansion," they conclude.