In-Depth

IBM's Mainframe Feature Set Trickles Down -- And Up

zSeries used to the be the source of features for other products in IBM's line. Now it may be the other way around.

• By Stephen Swoyer
• 05/13/2004

This in itself isn’t newsworthy, of course: Since reorganizing its disparate computing assets under the eServer brand almost four years ago, Big Blue has steadily migrated features from zSeries on down. But last month’s announcement highlighted another trend, one that’s the reverse of IBM’s mainframe trickle-down: Technologies are migrating north from zSeries’ lesser siblings. While analysts expect that the mainframe will continue to be IBM’s tier one platform for many future technology improvements, they note that Big Iron could increasingly be on the receiving end of a few improvements itself.

Gordon Haff, a senior analyst with consultancy Illuminata, says that IBM’s eServer product shuffle—prompted in late 2000 by intense pressure from Sun Microsystems Inc. and Hewlett-Packard Co.—has been an unqualified blessing for Big Blue. “The way groups within IBM worked together, sort of pre-eServer, the different product groups were more likely to throw things at each other than they were to share technology, and that’s just totally changed,” he points out. “So, absolutely, the eServer unification is a big part of enabling not just the technology-sharing, but … really enabling all of the various flavors of eServer to really emerge and be very successful.”

But the Virtualization Engine IBM announced last month marked a new stage in technology trickle-down, says Tim Dougherty, director of eServer products for IBM’s systems and technology group. “We’ve been doing [technology-sharing] for a number of years, but a lot of that was just the concepts. Here we’re actually taking code bases that we used in the mainframe and bringing it down into pSeries and iSeries,” he points out. “Similarly, we’re taking the code base that we used on xSeries and moving it up.”

For example, Dougherty points out, although IBM has supported logical partitions (LPAR) on both pSeries and iSeries for quite some time, the concept of logical partitioning was first developed on Big Blue’s mainframe systems. In the same way, IBM first introduced support for micro-partitioning—i.e., the ability to run hundreds or even thousands of LPARs on a single processor—on Big Iron. Virtualization Engine brings those capabilities to Big Blue’s i5 and pSeries systems, says Dougherty, and is based (at least in part) on the same code IBM developed to support micro-partitioning on its mainframes.

Similarly, high-availability, data replication, and other high-end capabilities—such as the ability to concurrently host several operating environments on a single i5 or pSeries system—have migrated southward over time, from both S/390 and zSeries. Now that IBM has shifted its focus to so-called On Demand computing, analysts and industry watchers expect that this trend will continue.

Indeed, to the extent that IBM’s On Demand computing model describes an enterprise-computing infrastructure that is both highly flexible and efficient, it could have no better proof of concept than the zSeries mainframe. “Over time … the intent is to drive z[Series] levels of capability everywhere,” confirms Jonathan Eunice, a principal analyst with Illuminata, who notes that Big Blue’s i5 and pSeries platforms should continue to trail zSeries for at least the next five years—on the partitioning front, at least.

As IBM attempts to incorporate On Demand features into its wild profusion of existing products, it’s pushing highly esoteric technologies—such as automatic processor balancing—into other eServer lines. “zSeries has a function that we have that we call automoatic processor balancing. This is a capability that allows you to automatically adjust system performances, so if you have several LPARS running on a single processor, they can share a core of resources underneath, and the LPAR that needs [the resources] most are automatically given access,” explains Ian Jarman, product marketing manager for IBM’s i5 servers. “This is a capability that drives up the utilization of a processor, so we’re raising utilization using these virtualization capabilities.”

Similarly, other mainframe capabilities, such as Big Blue’s Capacity on Demand (COD), which lets users unlock additional system capacity to accommodate increased demands, have also found their way into iSeries and pSeries.

In some cases, the fruits of mainframe technology trickle-down have been realized behind the scenes --- or under the covers, as it were. Dougherty cites the example of IBM’s eServer BladeCenter server blades, which were among the first to ship with higher densities (i.e., greater than uniprocessor) and other advanced features, such as fibre channel connectivity (see http://www.esj.com/news/article.asp?editorialsId=281). “The power, packaging and cooling that we do around BladeCenter was actually done by what used to be the zSeries power, packaging, and cooling team, which is the reason that we have been able to fit so many processors into such a small form-factor,” he observes. “Prior to [the eServer reorganization], this was a team that used to work exclusively on the mainframe.”

Quid Pro Quo

After years of prolific technology seeding, the mainframe is finally collecting its rewards. Take COD, which originated on the mainframe and which was widely disseminated to Big Blue’s other platforms. A great idea, to be sure, but not without a catch or two. For example, once customers unlocked capacity, they couldn’t turn it off again. In effect, as soon as they unlocked it, they bought it. What was needed, customers told IBM, was a “light switch” capability such that they could turn extra capacity on and off to meet changing business conditions.

But when IBM finally introduced its light-switch COD capability, called, appropriately enough, On/Off Capacity on Demand (COD), it did so first on iSeries, and not on zSeries—in spite of the fact that On/Off COD is a godsend for mainframe environments, where capacity is massaged to the last MIP.

On a related note, the Director Multiplatform software that IBM introduced last month in tandem with its Virtualization Engine technology was originally developed for its xSeries systems. IBM now positions it as a management facility for the breadth of its eServer line, along with non-eServer systems. Much of the conceptual work on which Director is based originated in the mainframe space, says Dougherty, and now the chicken has come home to roost, so to speak. “It’s a culmination of three plus years of work, but a lot of the research work is primarily in this area of enterprise workload management, of which a lot of that functionality comes from the mainframe,” he confirms.

Over time, says Illuminata’s Haff, the mainframe should continue to give as much, if not more, than it gets. Nevertheless, he suggests, the disparity between give and take should continue to narrow. “You’ll certainly see a continuance of trickle down, but I think it’s probably simplistic and it will be increasingly simplistic to sort of think of it as this pure waterfall coming down from pSeries to iSeries to xSeries, to blades,” he asserts. “You have different demands for different sets of products—different priorities—so there will tend to be development first at the high-end. On the other hand, high-end pSeries and iSeries are pretty large systems these days, too, so you’ll tend to have a trickle down effect, but there are going to be sort of a lot of cross-sharing from a lot of different directions.”