Progress Is Our Most Important Product
March 31, 2008 Timothy Prickett Morgan
In the 1960s, when General Electric was still a maker of computers in addition to lots of other gadgetry, the company had a catchy slogan, one perhaps even more famous than IBM‘s “Think” admonition from its stern founder, Thomas Watson. That GE slogan–“”–has become as much of the modern business lexicon as the suggestion that we should think outside of boxes or that team does not have an “i” in it.
This is all particularly ironic to me as I sit here on a cloudy New York Friday afternoon, mulling over IBM’s tactics and strategies for the OS/400 and i5/OS platform for the past 20 years as I prepare to pack up my laptop and head down to Nashville, Tennessee, to attend the COMMON midrange user group meeting. The AS/400, if you still insist on calling it that, is coming up on its 20th birthday, and as readers of this newsletter know, I have a hunch, as I wrote about a few weeks ago, that IBM will kill off the System i and System p brands and come up with some single brand–probably Power Systems, given the name of the division that is designing and making Power-based servers within the Systems and Technology Group.
We have come a long way since the AS/400 was launched as the merger of the System/36 and System/38 platforms back in June 1988. In a lot of different ways we have come a long way. And the people who have grown along with the platform, stuck by it, defended it, and urged IBM to keep changing and improving it are one of the reasons why the platform is still around two decades later when so many others have fallen by the wayside.
Just to give you a visual to encapsulate the magnitude of the changes we have absorbed, I updated a chart I made for The Four Hundred on the eve of the launch of the “Northstar” PowerPC 6XX and 7XX line back in 1997. At the time, I did projections for performance out to 2002, and I will tell you that IBM hit my near-term projections in the late 1990s, and then with the Power4 generation, it started to exceed them. The progress we have seen, in terms of performance and bang for the buck, has been substantial. Take a gander at this:
This chart shows the most powerful System/38, AS/400, iSeries, System i, and Power Systems server from January 1988 through December 2008. By my estimation, the System/38 Model 700 had a Commercial Performance Workload (CPW) rating of around 8, and at a cost of $192,000 for a base system, the cost per CPW for this box–which was more expensive by an order of magnitude based on performance than a mainframe at the time–came to over $24,000. The launch of the B series AS/400s–the B60 in 1988 and the faster B70 in early 1989–cut that cost per CPW in half. Since 1991, when IBM launched its first two-way D80 machines, which put two CISC processors in a shared memory symmetric multiprocessor configuration, IBM has used a combination of SMP clustering, faster processors, and in later years, multiple cores per processor socket, to keep the performance of the machine pushing ahead.
The chart above shows this progress on a log scale, which is why it looks fairly linear. The data is choppy because IBM has not made system changes every quarter, but roughly once a year, sometimes once every 18 to 24 months. The chart also shows how important SMP clustering has been to IBM, and the scalability that IBM has brought to bear, packing 24, 32, and 64 cores in a single system image, has given OS/400 and then i5/OS a lot of room to grow. A decade ago, no one would have ever expected IBM to go this route, but then again, we were only thinking about jacking up clock speeds back then, not power and cooling issues. The reason we have what are in essence baby SMPs inside a chip socket these days is not because this is a great way to build computers. It is because there is no other way to ride Moore’s Law and not get fried, and as I have said before, sooner or later this multicore approach will run out of gas. IBM seems to get this, and that is why it rejiggered the instruction pipeline the Power6 chips, allowing it to possible push clock speeds as high as 6 GHz eventually, thereby making application threads run faster rather than trying to run more application threads at the same time. This shift in instruction pipelines came at a cost, however, and I don’t think the forthcoming Power6-based System i 595 machine–or what we would have called a System i 595–will pack more than about 325,000 CPWs of performance. That is still not too bad compared to the 216,000 CPWs of the Power5+ version of the System i 595, but it is not the 450,000 CPWs or so you might guess based on moving the clock speeds from 2.3 GHz with the Power5+ multichip modules (MCMs) up to around 5 GHz with the future Power6 595s. But the changes IBM has made will allow it to push up to 6 GHz, I believe, and therefore get another 20 percent more out of the box and still stay within its thermal envelope.
Beyond that, it is hard to say what Power6+ has in store, or Power7 and Power8, for that matter. But perhaps IBM could have called these machines Progress Systems, with power getting some negative connotations these days; progress is the one thing we can count on from Big Blue. As I pointed out a year ago in a preview of Power6 machines and some discussion on Power7, IBM was debating moving up from 64-way SMP to 128-way SMP with the Power6+ machines, and is expected to use the 45 nanometer shrink in 2010 or so with the Power7 generation to squeeze up to four cores in a processor socket.
No matter how IBM does it, what I know for sure is that Big Blue will not make the mistake it made with the System/38 and early AS/400s, which had workloads that outstripped the performance that IBM could deliver. IBM will have boxes so powerful that very few customers will ever come close to hitting the performance ceiling. Whether IBM packages up that performance correctly and prices it fairly is another matter, of course. And it is here, more than any other area, where IBM has needed to make progress for the better part of the decade. The price-to-value ideas embodied in the user-priced System i 515 and 525 machines and the experimental and evolving pricing for the System i 570 gives me hope that IBM understands that it needs to stop discriminating against System i customers with excessive processor, memory, disk, and software prices.
I remain hopeful. Change is always possible. It is how we get progress, after all. One step at a time.