Where the iSeries Meets the Xbox
May 17, 2004 David Morris
The spring COMMON conference was full of surprise announcements from IBM. As you know by now, IBM is delivering Power5 technology on the iSeries that is 18 to 24 months ahead of anything Intel has to offer in its X86 chips. Another surprise came from IBM’s chief iSeries scientist, Frank Soltis, who outlined plans for Power. Soltis described how IBM has been working with Microsoft to deliver Power chips in the next generation of its Xbox game machine.
An evening presentation by COMMON and IBM for local user groups started off with Soltis telling stories about his children, as any father is apt to do. Soltis described a banking customer located on the edge of a straw market in Canada South. (Within IBM, the term Canada South refers to the Caribbean islands, which are supported by IBM Canada and attract IBM salespeople and scientists in the winter.) Apparently, a fire broke out in the straw market in front of the bank, and the local fire department flooded the bank and its iSeries system in an effort to hold back the flames. The bank’s backups were incomplete, and IBM was brought in to recover what they could from the wet disk drives. It turns out hair dryers were all the IBM recovery team needed to retrieve the bank’s data.
Soltis then relayed another story to underscore the difference between pSeries Unix and iSeries customers. This time, Soltis was called in to meet with a major customer that was considering investing in a new Unix platform. He said Sun Microsystems had sent out a lead scientist who was an Ozzie Osborne look-alike and impressed the customer by talking about Sun’s Level 2 cache architecture. When Soltis met with this customer, the discussion focused on the relative merits of Sparc and Power4 processor cache sizes and algorithms, which is something most iSeries shops don’t care about. In this case, IBM won and sold the customer on the pSeries.
IBM’s new i5 and Power5 chips are an impressive combination, and it’s been a long time since iSeries customers have led with new technology. During the past few years, IBM has let the pSeries lead the iSeries in announcing support for new hardware, so you might conclude that IBM’s decision to support AIX on the i5 will lead to a convergence of the iSeries and pSeries products. Soltis downplayed this possibility repeatedly, pointing out that pSeries customers care about cache sizes, chip speeds, and MIPS, while iSeries customers care about ease of use and reliability, not about how fast their processor is.
What can be honestly said is that IBM is converging the hardware to squeeze more profits out of the Power boxes, but it will continue to package them as separate products for OS/400 and AIX customers with iSeries and pSeries (soon to be i5 and p5) servers, even as both platforms support each other’s operating systems.
Another first with the i5 announcement was IBM’s decision to put the latest Power5 chips into low-end Model 520 systems. In the past, IBM has only released its latest chips into the midrange and high-end Power servers in the iSeries line. Soltis and other IBMers say that is because Power4 chips ran too hot to be in entry servers. While this may be the iSeries team’s argument for not using the Power4 and Power4+ chips, witness the two-way pSeries 615 and the four-way pSeries 630, which are entry machines that use these chips. The iSeries team made a choice to stick with the S-Star PowerPC processors at the low end of the iSeries line, and maybe for good reason. But it wasn’t possible to make an entry Power4 or Power4+ server, since IBM was already doing it in the pSeries line. That said, the Power5 processors have technology that definitely makes them a better option for entry machines. The Power5 chips use a new technology that shuts down parts of the processor that are not being used by workloads, thus reducing the amount of heat that is generated.
The ability to run chips cooler is critical to building faster processors. Soltis claims that Intel has been unable to get its Itanium chips to run cool and that Intel won’t have an answer to heating problems for several years. (In the two weeks since Soltis gave his speech at COMMON, Intel has decided to move to dual- and quad-core Xeon and Itanium designs that will run at much lower clock speeds and therefore dissipate less heat. Sound familiar?) Intel has been pushing up the clock speed on Xeon and Itanium processors, but this approach has reached a point of diminishing return.
Another problem for Intel, says Soltis, is Microsoft’s decision to use three Power5 chips in its Xbox game systems. To run that type of system, you need a pretty sophisticated operating system. This simple idea has lead to speculation that Microsoft will adapt some form of its upcoming “Longhorn” kicker to the current Windows XP/2003 operating system to control these game machines.
Longhorn will be a true 64-bit operating system and feature something akin to the single-level storage developed for the IBM midrange decades ago. Windows NT also has a hardware abstraction layer very similar to the technology-independent machine interface used in the AS/400 and iSeries. Soltis, who is considered the father of IBM’s System 38, which was the first computer to have single-level storage and a hardware abstraction layer, back in 1978, talked about how Power6 and Longhorn, which Microsoft plans to deliver in 2007, are a perfect fit. Longhorn could eventually become part of IBM’s virtualization strategy.
Soltis loves to talk about chips, and this evening was no exception. He described where IBM is today and where it is heading. The Power5 chips announced at COMMON run at 1.5 GHz and 1.65 GHz and will run as high as 2 GHz; they are built using a 130 nanometer process. In about nine months, Power5+ will be announced and will use a 90 nanometer process. In about 18 months, IBM will move on to Power6, which will use a 65 nanometer process (that is 4 atoms) and run at up to 5 GHz. IBM is testing Power6 chips in the lab right now, incidentally.
After Power6, IBM will continue to release new versions every eighteen months or so. IBM’s current roadmap goes out to Power9, which should arrive in about six years. Each generation will get faster and networks will increasingly become the bottleneck. And the so-called “Virtualization Engine” at the heart of the Power series of servers will allow customers to put hundreds and maybe eventually thousands of virtual computers on one physical server, minimizing network latency and allowing high speed processors to work more efficiently.
While this all sounds wonderful, as an OS/400 coder, I left this meeting wondering how OS/400 shops will use all of this computing capacity. Looking back, I realize that we will figure out something. Companies always do.