Soltis and Friends Give Their Visions for the iSeries
by Joe Hertvik
Frank Soltis is the chief architect of the iSeries line, and he is not afraid of controversy or of speaking his mind on whatever issue it might be that affects the product line that he, among dozens of others, helped bring to life 30 years ago. In the past decade, as the AS/400 and iSeries line came under intense competitive pressure, IBM has put forth Soltis as the key visionary and mouthpiece for the technology used in the iSeries line. Soltis took the stage once again at COMMON last week.
People always listen to what he has to say even though IBM itself sometimes does not. Soltis has been formally involved in IBM's midrange product strategy since 1983, so he has seen a lot of ideas and products come and go. His COMMON session this time around--called A Vision for the iSeries--talked about the future direction of the iSeries product line. Soltis' session is always one of the big draws of the show; in fact, it is so well attended that COMMON does not schedule any other sessions against it.
This year, Soltis' vision talk was followed a day later by a panel discussion with fellow IBMers Dave Boucher, a former OS/400 software strategist, and Carol Egan, who has also been deeply involved in OS/400 strategy. The panel presented their opinions on the iSeries product line and where it might be going in the long-term, roughly the next five to 10 years. It also provided IBM with a forum for customer feedback on where i5 users need the product to go.
It's important to understand that any statements that Soltis, Boucher, and Egan made as part of these discussions are not pre-announcements of future specific iSeries capabilities or product directions. These sessions provided an overview of the current state of the product line combined with a review of current trends and where those trends could lead to in the future. Trying to discern IBM's specific future moves with the iSeries line from what it says at such sessions is a mistake. Many iSeries product decisions have not yet been made and market conditions and IBM politics could fundamentally alter any plans that seem solid today when we get to five or 10 years down the road. That said, the sessions are a good window into IBM 's thoughts about the future, as well as providing some interesting information about past decisions.
A major part of Soltis' A Vision for iSeries session dealt with the processor performance gap between IBM Power processors and Intel's competing Xeon and Itanium architectures. Soltis talked about some limits of Moore's Law, which declares that because of advances in chip manufacturing technologies, processor performance can double roughly every 18 months. You make smaller circuits and you can make them run faster. This has been the Intel design philosophy for three decades and an idea that Gordon Moore, one of Intel's founders, put forth 40 years ago. However, this approach to chip making has its limits, and they all have to do with heat. Once you get above 5 GHz or so, chips get so hot you really can't use them in general purpose, air-cooled computers. So now, Intel and its server and PC software partners can no longer count on a steady diet of increasingly faster X86 processors to make single-threaded applications run quicker.
IBM took another route away from dependence on Moore's Law with the Power processor line, Soltis says. The company introduced many new processor features in the Power line, like SOI, multi-core processors, and other tricks to make existing processors run faster in a parallel environment. With Power, IBM feels it doesn't need to double its clock speed to double performance. You can run multiple applications at the same time, which isn't as feasible or easy in the Intel world. At least not yet.
Soltis did eventually point out is that Intel has seen the error of its ways and will be rolling out its own dual-core processors. As we have reported in other IT Jungle publications, these dual-core X86 chips will roll out over the next nine months, starting with dual-core desktop Pentium processors, moving up to dual-core Xeon and Itanium server processors, and ending with dual-core Pentium M/Centrino laptop processors. Intel, says Soltis, is now pursuing many of the same breakthroughs that IBM has had for years, starting with the AS/400 back in the 1990s. Soltis advanced the notion that, to its detriment, the software industry is dependent on a 35-year drive to increase clock speed. This hurts Windows developers because their single-threaded applications will not run any faster on a multicore chip. The industry is in for a shock, he said, and people are starting to realize that a lot of Intel applications will need to be rewritten.
What Soltis didn't say is that in the Windows server environment, the operating system and much of the Microsoft middleware stack has been enabled to support Intel's HyperThreading (HT) simultaneous multithreading, which has been available for two years, and that any software that has been enabled for HT (whether it is on a desktop or on a server) will be able to make use of multicore X86 processors without any further application rewriting. And for those applications that are single threaded (just as is the case for the vast majority of homegrown OS/400 applications, by the way), the ability to multitask applications across many cores has its own benefits. Intel is clearly behind IBM--there is no question about that. But so was the entire industry, and the entire industry is catching up, at least on this front.
The iSeries and i5 line benefit from Power's advantage over Intel, says Soltis, because the Power architecture that produces sky-high benchmarks on pSeries boxes is also running underneath the iSeries. And not only can you run OS/400 on an i5 or iSeries box, you can also run AIX, Linux, and even single-threaded Windows applications on the Integrated xSeries Server or outboard xSeries boxes attached to the iSeries through the Integrtaed xSeries Adapter. IBM introduced the dual-core Power5 processors across the entire i5 and p5 product lines last year, and the iSeries product line benefited by getting the technology first, he said. (How much of a benefit the iSeries gets from being first with Power5 or Power6 remains to be seen.)
Soltis also noted how the rest of the industry is starting to notice the Power processors. In the last six months, IBM has gotten more aggressive with Power processor technology, mostly because it sold its PC business off to Lenovo Group late last year. Big Blue is almost to the point of challenging Intel, and IBM would not have been able to do that if it was still dependent on Intel for technology or chips for its desktops. Selling its money-losing PC division eliminated that dependency. However, IBM is not totally weaned off the Intel iron. In most quarters these days, it generates more revenue from sales of xSeries servers than any other of its eServer platforms. (However, it also probably generates the least amount of profits from the xSeries line.)
As we have previously reported in this newsletter (see "Future 'Cell' Power Processors Can Run OS/400", IBM, Sony, and Toshiba recently announced their new "Cell" chip, which will soon be the at the heart of workstations, high definition TVs, PlayStations, and other electronic equipment. Sony says the "Cell" chip will be a thousand times faster than its current PlayStation chips. And, because it's a Power processor, the "Cell" will be capable of running OS/400. It's also worth noting that IBM Rochester is involved in designing the "Cell" chip, which might lead to other i5 benefits in the future. (It is also worth noting that it is not a foregone conclusion that all technically excellent PowerPC or Power chips can run OS/400. The 64-bit PowerPC 970 chips used in Apple Macs and Xserves and in IBM's own BladeCenter blade servers cannot run OS/400.)
As Soltis pointed out last year, Microsoft has also selected a variant of the Power processor for its next generation of Xbox machines, which also offers the possibility of running i5OS on that platform. And as we have been saying for quite some time here at IT Jungle, since the Xbox is a 64-bit Power processor with a hardware abstraction layer that will allow it to run a variant of Microsoft's Windows software. If you can run Windows on the Xbox, then it is a small step to running Windows on the iSeries. After all, Soltis noted that IBM has been in discussions with Microsoft for the last four years to run Windows on Power. And the top brass of Sun Microsystems, which recently launched an open source variant of its Solaris platform (appropriately called OpenSolaris), say that they are considering porting Solaris to the Power processor--or letting the open source community do the dirty work. If this happened, Solaris could also run on an i5 box. Soltis also mentioned that modern Power processors could be running inside automobiles that are increasingly using multi-processor systems; the Motorola 68K family of processors that is the predecessor of the Power line was the king of the embedded computing world (which is why IBM chose them for its AS/400 IOPs all those years ago) and the Power chips from IBM and Motorola have largely supplanted the 68Ks.
There's also the BlueGene/L parallel Linux supercomputer that IBM is building for the Department of Energy at Lawrence Livermore Labs (which we have covered extensively in The Linux Beacon). When delivered over the next 18 months, this one system will contain half of the supercomputing processing power in the world--360 teraflops. The final product will contain over one million processors, which means that Big Blue is designing in a lot of self-healing capability into the package. And since it's being designed and run in Rochester, this self-healing technology could find its way into the iSeries.
Soltis stressed that IBM is excited about rolling out all this Power technology. And IBM is delivering it to multiple worlds, including iSeries, Windows, xSeries, and pSeries. With the game systems and new technologies, it's even being introduced to kids entering school. The Power processor is a big opportunity, and as IBM continues to roll out their vision of an eServer box that runs several different operating systems under the same hood, it's good to remember that a true eServer box is basically called an iSeries.
In other areas, Soltis mentioned that the next release of the i5/OS may be able to run an older version of the operating system, but it was unclear which version(s) he was referring to. Soltis also discussed extending the i5 and Power processing capabilities by "borrowing" power from outside the box, which is essentially the grid computing project that IBM continues to work on. He mentioned that we should continue to watch the grid space.
Panel Vision Sheds Light on Various Topics
As opposed to Soltis' speech, which was more focused on the benefit of Power technology, the panel he hosted tackled a number of different topics, based on questions from the audience. Here are some of the more interesting comments that came out of that session. Remember, however, that the topics discussed here are not hard and fast directions or announcements from IBM. Many of these items should be thought of as informed opinions instead of things that will definitely come to pass.
When asked whether there would be new partition options in i5/OS for running IBM mainframe software, the panel said that this was a marketing issue, not a technical issue. They also noted that IBM used to get questions like this about AIX. Mainframe compatibility has been discussed inside IBM, and some of the original design points for the Power6 processor may have included mainframe partitions, but those points may have also been eliminated from the spec as it was developed. (This is the so-called "Project ECLipz," which we wrote about in The Four Hundred a year and a half ago.)
As far as loading Windows in a logical partition, Soltis said that this is a business decision that will be made by Microsoft, and IBM really has very little to say in whether that happens or not. What may drive Windows onto Power is that the processor industry has become more of a two-horse race between Power and 64-bit X86, whether those X86 chips come from Intel or its rival, Advanced Micro Devices. Because of that, Power is the only new architecture that Microsoft might be interested in looking at. But don't forget that the next generation of Xbox machines will be based on Power processors, so Microsoft has already done a lot of the work. The issue is how much heat can Microsoft take from Intel? (The answer, by the way, is tricky. How fast do you think Intel would start pumping money into Linux to tune it for X86? And IBM would have to counter with similar positioning for Linux, which would irritate Microsoft.)
When asked about the consolidation of packaged solutions in the software industry, where one vendor buys several products in the same segment, the panel noted that they don't necessarily see vendors always consolidating these different packages into a new, better package. Morphing applications isn't always in the vendor's best interest, unless there's some underlying business justification. But while saying that, the panel also acknowledged the power of the open source community where several open source packages, such as JBoss and Linux, are starting to creep into the business environment. Open source could represent a sea change looming on the horizon for packaged software, with new open source options appearing at the same time that the software market consolidates ownership of application segment solutions.
As far as hardware configuration changes go, the panel saw a good possibility that i5 storage may migrate out of the box entirely in the next five years. The rest of the industry is going in this direction, and it may just be a matter of time before IBM moves to standard storage on an i5.
Regarding what IBM is doing to recapture market share in the midrange market, the panel mentioned that there are some discussions in IBM on what the company can do to replace older smaller machines in dormant accounts and move those customers up to newer hardware. This would be similar to the System/36 replacement project IBM undertook in the 1990s, where the company offered customers such an attractive offer that they would have been crazy to refuse it.
IBM has also boosted its direct iSeries sales force. Most of the new people are focused on selling new accounts. The panel said that IBM also has launched an effort to work with ISVs who are not currently supporting the iSeries and to port those applications to the iSeries. Some of this may grow out of the effort to move X86-Linux applications to Power. IBM is funding work to identify X86-Linux applications on the list, and as we have previously reported, has created a set of tools called "Chiphopper" to make it easier to code once for any Linux application across the entire eServer line.
The panel said that IBM plans to provide more enhancements to the cumulative PTF process. IBM is planning on doing more PTF fix automation but there are a number of items to be worked, including how customers will receive automatic fixes and whether they will allow the fixes to be automatically installed; does IBM include AIX and Linux fixes in the PTF process; and whether OS/400 should emulate the Microsoft software fix model more.
The panel also discussed the future of the DB2/UDB database technology. Panelist Dave Boucher did not believe that there was going to be a new paradigm shift in OS/400 database technology to replace the relational database model. He didn't see any clear emerging technology, but he did see a need for our systems to support bigger and bigger databases.
As far as software and system delivery, the panel did mention that they see a trend toward the customer downloading and installing more new software packages themselves, and away from IBM pre-loading software onto hardware at the factory.
When asked whether they saw enrollment in computer science programs in colleges going down, their response was that enrollment was definitely down but it may be due to a valley in a 10-year cycle caused by the dot-com bust. Robotics is a potentially hot new area that hasn't taken off yet, while operating system development has now become more of an incremental process.