What The Future Holds For IBM i Platforms In 2013
January 7, 2013 Timothy Prickett Morgan
The future is never wide open, but rather is a sphere radiating out from this moment at precisely the speed of light. Everything that can and will happen will occur within that sphere, events moving out from the point where you sit and events moving in from outside, from all directions and as if they are pre-coordinated from the past outside of you to affect you precisely, right here, right now, from the edge of that sphere inwards. When you think about it that way, the orchestration of all of our presence is all the more remarkable. Outside of that spacetime bubble, it may as well be void for all that it can affect you.
Well, unless there are wormholes that can roll up space and stretch out time, and given the immensity of the universe, I sure hope there are. Just not too close to where I am sitting, trying to drink my second cup of coffee on the morning of 1/1/2013, writing this essay, gazing into my internal crystal ball to try to figure out what we in the IBM i community might be facing in this year that is already unfolding.
The obvious thing that is coming our way, and almost certainly as fast as Big Blue can manage it, is the full rollout of Power7+ systems. IBM started talking about the features in the Power7+ processors, which shrink down to 32 nanometer processes and allow for a huge boost in L3 cache memory and a modest bump in clock speeds, back in August. The first machines to feature the Power7+ chip, the multi-chassis Power770+ and Power 780+ machines, came out in October, and then in late November, when no one was expecting it, the Power7+ was dropped into the Flex System p260+ server node for the PureSystems converged modular system.
We know that the Power 795 is not going to get a Power7+ bump, since IBM told us this back in October last year, and this is consistent with IBM’s past practices. (Although, as I have said, if a big Power 795 shop comes to IBM asking for a Power7+ upgrade, I am fairly certain the company could make one. With all of those accelerators to speed up memory compression, hashing, and security on the chip and the extra L3 cache memory boosted from 32 MB with the Power7 chips implemented in 45 nanometers to 80 MB with the Power7+ implemented in 32 nanometers.) The random number generator accelerator might be something big Power Systems shops don’t want to wait for until the Power8 generation gets underway, perhaps in the middle of 2014 or so.
I have no doubt that the Power7+ chip will eventually be plunked into a four-way p460+ server node for the Flex System chassis, and I think there is even a fair chance that a single-node p160+ server node will eventually be launched, perhaps with only four of eight cores available and using partial-dud chips as IBM currently does in its PS7XX blades and the Power 720 rack and tower servers. This would be a low-cost Flex System node, aimed more at those who are budget conscious than those looking for the maximum amount of compute per rack. I obviously think there will be Power7+ updates for the Power 710, 720, 730, 740, and 750 machines this year, and as soon as IBM can swing higher volumes of the latest Power chip. The Power720+ is the key box for IBM i shops, with the Power 750+ coming in right behind it.
What I am not sure of is this: Why IBM needs to continue expanding the BladeCenter product line. The compute density of the Flex System chassis is a little better than with the BladeCenter, and the server nodes are designed to be able to take regular processors and memory and don’t have to be flattened like a pancake as in the BladeCenter blades. If I was shopping for a modular machine today and I was buying from IBM, I would not invest in the BladeCenter and would invest in the Flex System. This is clearly IBM’s future, and while Big Blue will sell and support BladeCenter machines for many years to come, that does not necessarily mean that it will put a lot of effort into enhancing the BladeCenter product. If margins get tight, and IBM has to cut somewhere, this is the obvious place to pull back. IBM has already skipped several generations of Opteron processors from Advanced Micro Devices with the BladeCenter products, so there is already some precedent here. IBM can’t do this with its Xeon-based blades, of course, since the Intel products represent the bulk of its blade installed base.
But IBM could certainly freeze-dry the Power-based blades as they are and not affect its sales too much. As we all know, Power-based blades never were a hot seller in the OS/400 and IBM i market anyway, and that’s a shame given the integrated nature of the BladeCenter platform and the operating system alike. BladeCenter boxes just didn’t have enough native software support and attractive enough pricing to take off and make a dent in Power 7XX rack and tower system sales. And even if IBM does do a Power7+ revamp with the BladeCenters, which would be easy enough considering the socket compatibility of the Power7 and Power7+ processors, I can say with a reasonable amount of certainty that I do not expect for there to be BladeCenter blade servers based on Power8 processors. Modular is the way of the future, and if the Flex System machines take off as IBM hopes, there may not even be standalone rack servers, either. IBM doesn’t offer multiple mainframe form factors, and I think it will not want to do multiple Power form factors years hence. The Flex System approach gives you all of the benefits of rack and blade machines in one box, so what is the point?
Cloud, cloud, cloud
One more thing about the impending double whammy Power7+ processors, which will cram two eight-core chips into a single Power CPU socket. While I would love for this chip to be available for IBM i platforms where threads and cache are more important than clock speeds, I think there is a fair chance that this will not be a general purpose chip.
In fact, I think that IBM will use the double-stuffed Power7+ as a stop-gap against the Xeon onslaught in the data center and perhaps only put it into PowerLinux machines, which are designed to run only Linux and not AIX or IBM i. I say this with one caveat: IBM may allow the double-stuffed Power7+ machines to run AIX, IBM i, or Linux at service providers building clouds. In this case, being able to put as many cores into a box and pinning a customer to a core is what makes the business case for clouds. In this way, IBM can use whatever aggressive pricing it wants for AIX and IBM i cloud builders and not affect the revenue stream coming out of its generic IBM i installed base.
And speaking of clouds, this might be the year that true clouds–meaning on-off capacity backed by a cloud controller fabric, not just a hosted logical machine slice–takes off on Power-based systems. The deeply discounted hardware and systems software deal aimed at managed service providers that IBM debuted as 2012 was winding down is the first big step toward making clouds economical for MSPs to build. The cheap hardware, at 55 percent to 65 percent off list price, and quarterly IBM i and AIX licensing, which costs $500 per core over three months, makes it economically practical to build IBM i and AIX clouds. It has been technically possible to do it since a hypervisor was first available on OS/400 back in 1998, but upfront hardware and software costs made it very risky to build what we now call a cloud.
Maintenance hike, V5R4 deals likely
It has been a while since IBM hiked the maintenance on Power Systems iron, and with the economy a bit skittish but not technically in recession except in a few pockets of the globe and a transition under way in Power Systems iron, I think IBM will use a carrot-and-stick approach to encourage customers to move to new iron and away from old iron with relatively pricey monthly maintenance fees.
IBM last raised maintenance prices on Power-based systems in March 2011 and a slew of other iron, with price increases ranging from 3 percent to 5.4 percent and averaging around 3.4 percent across the machines. A much steeper maintenance increase ranging from 5 to 9.5 percent on OS/400 and i5/OS iron was instituted as the Great Recession was building up a good head of steam in March 2008, and an even steeper set of increases ranging from 9 to 19 percent on vintage iSeries and AS/400 boxes was slapped onto the maintenance bill in early 2006.
The thing about a pattern is that you have to know when it will break as much as when it will hold. With the economy doing more or less alright in the United States and in parts of Europe and Asia but server sales on the skids, you have to figure IBM wants to make up some sales with extra maintenance right now from customers who do not move to new iron. IBM waited three years between the last set of maintenance price hikes, but I think it will revert to a two-year gap now that we are through the worst of the Great Recession, and hence we should expect price hikes sometime in February, March, or April, and very likely in concert with the launch of entry and midrange Power7+ iron.
I also expect for IBM to start doing some wheeling and dealing to get customers to move off old iSeries and Power Systems iron running i5/OS V5R4, which loses its technical support on September 30 of this year. There is service extension support if you are desperate, which will be very pricey indeed, but at this point, IBM will be encouraging customers to do RPG and COBOL program conversion and move to IBM i 7.1 with its more flexible Technology Refresh updating.
I said it before and I will say it again: I think IBM should stop forcing these customers to do a program conversion and figure out how to run V5R4 applications in emulation mode atop the PowerVM hypervisor, perhaps in a V5R4 shell that is actually backed by IBM i 7.1 running the QuickTransit emulator, allowing that V5R4 application and database stack to pretend to be on older Power6 iron. I doubt very much if customers would care if this emulation offered terrible performance at first. They could get to new iron and gradually move their applications and databases to native mode, much as they could with System/36 and System/38 emulation back in the AS/400 days. This might be an expensive option for IBM to develop, and hence that might be why it has not been done even though IBM has owned QuickTransit for several years–and not done much useful with it except keeping it out of the hands of its enemies.
It is hard to say what kind of V5R4 migration deals IBM might offer, but a combination of hardware and software discounts plus porting services would seem to be in order.
If you have any prognostications to offer for 2013, we’re all ears. Send them on in and I will run them in next week’s issue.