IBM i Installed Base Dominated By Vintage Iron
December 9, 2013 Timothy Prickett Morgan
If the IBM i-Power Systems platform was up for an award for longevity, it would actually have to be given two awards. One trophy would be for persevering as the last platform standing from the Minicomputer Revolution back in the late 1970s and throughout the 1980s. And the second medal would be for machines that stay in the field for an incredibly, almost unbelievably, long time.
We all know this instinctively and anecdotally, having heard many of the legends of the AS/400 and its progeny out in the field. Stories of machines being sheet-rocked behind a wall or stacked over with pallets may be an exaggeration, or may not. But these machines persist longer than most iron in the field, and it is remarkable as much as it is an opportunity for IBM and its reseller and software partners to start thinking outside of the box to boost their sales and profits while helping the vast installed base of vintage machines move ahead and get modern.
IBM has not done line-item financial reporting that broke out the IBM i and predecessor platforms for many years now, but every now and again we get a hint of whether sales are growing or not from whoever happens to be the general manager of the Power Systems division at the time. Sales were down during the Great Recession, as you might expect, and recovered with the launch of the Power7 chips in 2010 in the wake of the recession. The overall server market followed much the same curve, even the juggernaut X86 server platform that is dominated these days by Intel. (The recession and a vastly improved server product in early 2009 pretty much sealed the fate of Advanced Micro Devices, which had been gaining on Intel from 2003 through 2006. One bug and one chip delay, and AMD was on the skids in the data center.)
The analysts at Gartner, IDC, and Forrester may track the IBM i platform internally as part of their casing of the systems market, but they don’t talk publicly about it to wiseguys like me and they won’t tell you what they know unless you shell out some big-time cash. But we do catch wind of some real data every now and again, and from this, a persistent hack can create a model of what the installed base looks like.
IBM has made no secret of the fact that it wants IBM i shops to consolidate their Power-IBM i and X86-Windows platforms onto PureFlex converged platforms. The company launched a new PureFlex-IBM i bundle as part of the October 2013 announcements, and I told you all about the sales pitch that Big Blue is making to try to convince customers to move to the converged platform. As it turns out, that sales pitch is aimed square at the belly of the AS/400-iSeries-System i-IBM i installed base, although that might have not been clear at the time. The sales pitch showed a one-core Power 520 machine in a P10 class with 10 Dell PowerEdge servers with four cores each being consolidated onto a PureFlex p260+ node with four cores (one running IBM i) plus one x240 Xeon node with sixteen cores. The setup had a price tag of $124,038 and had a payback time of 21 months.
That kind of math is going to be gone over a lot in the next year and longer, I think. As part of its PureFlex push, IBM told some business partners the basic shape of a chunk of the installed base, and I caught wind of this data and have used it to build my own model of the complete OS/400, i5/OS, and IBM i installed base. This is not machine shipments over time, but my best educated guess at what the actual installed base looks like. Take a gander:
We know that there are over 150,000 unique customer accounts out there, because IBM told us that back in the summer of 2012 when Colin Parris was general manager of the Power Systems division. Each site has at least one machine, and I reckon that there are around 10,000 customers using high availability software to double up their boxes. I have a hard time believing in this virtual age that there are a lot of shops with dedicated machines for application development, but there could be another dark pool of vintage boxes lurking out there for development and tape backup that are not in the data above. This model is for machines doing productive work or being used as high availability backups. So this model has a little over 160,000 machines in the worldwide IBM i and predecessor platform installed base. These are the sales targets.
There are some interesting things to note. First, I am lumping the base and plus versions of each Power chip generation together. Architecturally, there is often not a big enough difference to look at them separately, and the biggest price/performance gains and largest changes in system design (including processor socket changes) come at the base Power chip changes, not at the pluses.
It is immediately obvious that the Power5 and Power5+ processors, which date from 2005 through 2007, dominate the installed base. In fact, if you do the math, machines using Power5 and Power5+ chips represent about half of the boxes installed on the globe, or just under 80,000 machines.
Many of these Power5 and Power5+ machines are as much as eight years old, which is twice the average four-year age of an X86 server deployed in the data center. X86 servers can run longer than four years, of course, but if you keep them on warranty and do the math on power, cooling, real estate, and other costs, a four-year lifecycle turns out to be the most economical upgrade cycle for most enterprise customers. (Intel, for instance, has precisely that cycle, and a third of its 50,000 machines used in its chip design operation are four years old and looking for an upgrade right about now.)
Another interesting thing: With the shift from Power5 and Power5+ to Power6 and (to a much lesser extent) Power6+ chips, the base shifted from being dominated by P10 machines to P05 machines. As you well know, the performance jump from Power5 to Power6 was quite large, and many shops could do a push-pull upgrade and drop down a software tier. The resulting savings in software costs from IBM were considerable, and have no doubt not made Software Group very happy. But this is the effect of Moore’s Law, and the trick is to make customers want to put more and more work on the platform and make it up in volume. The trouble is, I suspect, the transaction workloads on IBM i platforms are not growing as fast as Moore’s Law. They could, if the platform was priced more aggressively, but it isn’t, so that’s that.
At some point, Systems and Technology Group would do better if IBM i had the same low price on the P10 tier as on the P05 tier, much as IBM has stopped charging more for the P40, P50, and P60 tiers. IBM is sort-of doing this with hardware. The new PureFlex p260+ server node, which has two sockets and four cores, is in the P05 tier, and the p460 and p460+ are in the P10 tier instead of the P20 tier you would expect.
At this point, I estimate that Power6 and Power6+ machines represent a total of 16 percent of the installed base, or about 25,500 machines. Those on older Power6 boxes would seem to be the most likely to move if they have growing workloads, and they may just ride out the current Power7+ generation and see what Power8 chips hold for them in terms of price/performance. With an estimated 2.5X boost in performance coming over the Power7+ chips, the 12-core Power8 chips is probably worth waiting for next year if you can hold out. The 19,400 machines with Power7 and Power7+ chips are probably not going to be upgraded any time in the next several years, unless again they are at sites with rapidly growing workloads and sites that are putting more work on their IBM boxes. These companies do exist, but as the installed base chart above shows, the vast majority of customers must have very stable workloads or they would not be on such ancient machines.
I reckon that the base of machines using Power4 and Power4+ chips is a little larger than for Power7 and Power7+ processors, in this case 26,400 boxes. I am just dumbfounded that these boxes are still out there, doing useful work, but I am also kinda proud of that fact, too. They are a testament to IBM’s engineering and customer frugality. There is another 9,000 or so even older machines out there in the base using various PowerPC-AS “Star” family processors.
Don’t even get me going on the System/3X machines that are probably still lurking out there. . . .
Given this installed base, it begs the question: What does the community–meaning IBM, customers, and partners–do about it? What can be done, what should be done, and what will be done to move these companies forward? That is the subject for next week’s issue.