Q&A With Power Systems Top Brass, Part One
May 9, 2011 Timothy Prickett Morgan
Q&A With Power Systems Top Brass, Part One
The editorial team from The Four Hundred had a pow-wow with the top brass in the Power and z Systems division within IBM‘s converged Systems and Software Group. Tom Rosamilia, who was tapped to run both the Power and mainframe lines at IBM last August, had been running the mainframe line before that July 2010 reorganization and was in charge of the WebSphere line of middleware within Software Group before that.
Rosamilia brought along his lieutenants, Ian Jarman, Colin Parris, and Linda Grigoleit, and I brought along Alex Woodie and Dan Burger to sit at the round table. Jarman has been the product marketing manager for the AS/400, iSeries, and System i lines for many years and was tapped to be the manager of the entire Power Systems systems software lineup two years ago. Parris was in charge of the systems software development labs for Systems and Technology Group until last August, when he was named vice president and business line manager for the Power Systems business, working directly for Rosamilia. Grigoleit is the program manager in charge of the Power Systems Academic Initiative, which tries to build bridges between IBM and colleges and universities to make sure people have the Power Systems skills that companies need.
Because this is a long interview, we’re breaking it into two pieces. I did most of the yakking at the beginning along with Rosamilia and Jarman, but other people dove into the fray in the second half of the interview, which we will run next week.
We started off talking about the Watson question-answer machine, probably the hottest topic in systems at the moment.
Timothy Prickett Morgan: I was up at IBM Research to see it, and have talked to David Ferrucci, the lead on Watson, a few times, and was obviously very excited about it and somewhat terrified by the idea that this stuff actually works. I saw the medical diagnostics being done by the Columbia University team using Watson, and the Jeopardy! game was one thing, but that really drove it home. I think it is very, very interesting, and I keep thinking that it is a trick, or it shouldn’t work. People went nuts reading about Watson, and everyone is trying to figure out how does it change things. I am hoping it changes things in a good way, but I think there are going to be an awful lot of low-level lawyers eliminated soon as well as people who do the kind of analysis work that, well, we all do here in this room. Some of us may not have our jobs much longer because they can be replaced by a really good expert system.
Tom Rosamilia: I don’t know about job elimination. I think it is really an augmentation for large information bases that are difficult to digest. The medical field is an easy example, and you got to see that. Jeopardy! was somewhat of a contrived scenario in that in most practical scenarios we could be connected to the Internet, and we couldn’t be playing Jeopardy!, and we don’t have to answer a question in three seconds, and we would allow interaction. And so, if Watson only had a 30 percent confidence, maybe you supply some more information and you see what happens as the confidence level goes up. You can actually drill down and see why is Watson so confident, what document caused it to be so confident? And if it is a journal on cardiology, maybe it is a document that a doctor wants to read in addition to coming to the diagnosis he already has. You can actually see the body of evidence that was critical in coming to that recommendation.
The legal aspects of that are a lot more complicated. What is the liability on this? Someone will say, “The medical assistant told me is what it was.” But that is true in so many things.
Timothy Prickett Morgan: That’s true of the current practice of medicine. Anyway, Watson is interesting, but now I want to talk about the IBM i platform.
We got a gauge of how the business was going in the fourth quarter of last year and the first quarter of this year. We are all curious about how the IBM i operating system is doing in terms of shipments, upgrades, and what the base looks like. How many people are using OS/400 V5R3, how many are on V5R4? We know V5R4 support is probably going to run out in a year or so. How is the uptake of i 6.1? Do people tend to go in two steps to get to i 7.1, or are they doing it in one fell swoop?
Tom Rosamilia: We did see growth in the operating system as measured by revenue in the fourth quarter last year and in the first quarter this year. I don’t know that I would come to the conclusion that we are seeing a resurgence, but it is certainly a level of renewed interest. We are seeing interest in 5.4 to 6.1 upgrades, and in order to be on Power7, people know they have to be on i 6.1 or higher. People have asked us why we don’t just retrofit 5.4 so we wouldn’t require people to move up, and that’s kind of a mixed story in that the number of things we did in 6.1 was designed to take advantage of the Power7 capabilities. So going there would sort of be a toleration of, not an exploitation of, the Power7 systems. So we need people to move to 6.1 in order to take advantage of the features and functions of the Power7 architecture.
And so we have seen more folks moving. I got a lot of noise on this when I first got to the job last August from my sales team, asking why we can’t just get 5.4 running on Power7. And that noise level has really gone down.
Ian Jarman: We have really transitioned now to almost exclusively selling IBM i 6.1 or 7.1. We’re actually withdrawing marketing of 5.4 coincident with the withdrawal of our last Power6 servers in May. And it is typical in that situation for us, in our normal lifecycle, to support 5.4 for at least another year. We have made no announcement about the end of support for 5.4, but they should expect at least a year after the withdrawal of marketing.
People are really moving to Power7, and when they move to Power7, they want to take advantage of the new capabilities, including integration with external storage. Many of these capabilities rely on 6.1 and 7.1, so there’s a natural progression to these new releases.
Timothy Prickett Morgan: Are people going straight from V5R4 to 7.1, or do they do it in two steps?
Ian Jarman: I would say that the majority so far have been going into 6.1, but it is perfectly possible to go from 5.4 to 7.1. And when I talked with Pete Massiello, the president of COMMON and one of the people who has done a great deal of work upgrading people to these new releases, he said that he is now recommending people go directly to 7.1. So I think we’ll see more of that, and there is no reason you would do a transition to 7.1.
Timothy Prickett Morgan: If you have certified an application to run on 6.1, does it pretty much work on 7.1?
Ian Jarman: Pretty much. As you know, the move from 5.4 to 6.1 required retranslation, and it was a little more complicated than the normal release-to-release upgrade. And that is why all ISVs took great care in that certification. However, in comparison, 6.1 to 7.1 is very simple, and there are no barriers to moving between those releases. Typically, ISVs will certify, and the big ISVs have certified already on 7.1. So there is no barrier to moving to 7.1 and we would expect the push into 7.1 to be much more characteristic of a normal release-to-release upgrade.
Timothy Prickett Morgan: And for the i next release [almost certainly to be called IBM i 8.1], will this be a similar pattern, not another requirement for retranslation?
Tom Rosamilia: We don’t anticipate that.
Ian Jarman: We don’t anticipate any future change like that. Technically, the complication moving from 5.4 to 6.1 was that there were still some very old programs that did not have their observability maintained, and once a program has been on 6.1 or 7.1, that will not be a problem again.
Tom Rosamilia: The other thing that I would tell you is that the data shows, from a systems software perspective including PowerHA, PowerVM, and so on, we have seen a tremendous interest, growing much faster than the underlying hardware as the penetration rates grow. A lot of people don’t have these, and you don’t have to buy a new box to put on PowerHA or PowerVM.
So the systems software has been a great focus for us, and we have a dedicated sales force around this now, which we have never had before. And the growth rates for us in the third quarter, fourth quarter, and now the first quarter of this year have all been in the 30 percent range for the combination of all systems software, including the operating system. That’s AIX, IBM i, PowerHA, and PowerVM.
If you think about it, the operating systems tend to be lowering that, so the growth rates would be even higher if I separated out just PowerVM and PowerHA. In fact, they would be up in the 50 percent range. This is exactly what we are trying to do. We want our clients to understand the value of the systems software that unlocks a lot of the power of Power7. So we have people doing more virtualization and doing high availability, whether it is on IBM i or AIX–it’s all good.
Timothy Prickett Morgan: What’s the penetration for PowerVM and PowerHA? I know that Power Systems has a really good attach rate compared to X64 servers, for instance, especially on the larger Power Systems machines running OS/400 and IBM i. And then the AIX guys kicked it in with lower prices and better functionality. So where’s it at now? What is the typical Power 720 customer doing? Are almost all of them using PowerVM?
Ian Jarman: It varies by the size of the client and the size of the machine, obviously. And it also varies depending on whether you are talking about AIX or IBM i. In the Unix space, between Power5 and Power6, there was a factor of three increase in the adoption of PowerVM. If you look at Power7, on our high-end machines, the adoption of PowerVM is practically 100 percent.
Timothy Prickett Morgan: Which is what it was years ago on big servers running OS/400 and IBM i.
Ian Jarman: Right. The large users of IBM i are all using PowerVM. And when we ask the people here running it on smaller machines like Power 720s and 740s, it is much lower.
Timothy Prickett Morgan: Is it 50 percent? Is it 30 percent? If you look at the industry average for new server sales of all types, it is only 20 percent.
Tom Rosamilia: Not even. And while that number keeps growing, I don’t think our numbers on those classes of machines are much different from what you would find with VMware. It is growing, but there’s a lot left to do. And that is why the software numbers are outpacing the hardware numbers.
[During his keynote at COMMON, Rosamilia cited statistics from McKinsey & Co that said CPU utilization on servers averaged around 5 percent globally a few years ago, and has doubled to 10 percent of total CPU capacity. “If you are in marketing, that’s 100 percent growth, two times,” he said to big laughs. “We’ve got a long way to go.” Obviously, on X64 boxes, which do not have a subsystem architecture in their operating systems like IBM mainframe and OS/400 machines do, you need virtualization hypervisors to take the place of proper workload management so you can drive the utilization on servers up higher. The penetration of hypervisors on Windows machines, which have the worst workload management, should in my estimation eventually exceed that of IBM i and z/OS machines among SMB customers.]
Timothy Prickett Morgan: Are people generally virtualization Power Systems machines to put on multiple IBM i images, or are they adding AIX and Linux to machines to support infrastructure workloads, too? What does the spread look like now?
Ian Jarman: For IBM i clients, it is primarily to have multiple IBM i partitions.
Tom Rosamilia: And there is the same thing around AIX. I don’t know that we have that many heterogeneous AIX and IBM i environments. It is mostly just cranking out 150 or 200 images.
Timothy Prickett Morgan: The IBM mainframe is the exact opposite. You have a lot of sites that have a lot of Linux and a lot of MVS. I meant z/OS–I will never get that right.
Tom Rosamilia: On the mainframe side, we have around 4,000 customers worldwide. About a third of them–about 1,300 of them–have Linux somewhere in their enterprise. The vast majority of them have co-location of zOS and Linux on the mainframe. We actually seeing the emergence of a small group right now–around 30–who only have Linux and do not have any z/OS. We call that an Enterprise Linux Server, and it is starting to grow and we are seeing increased interest based on the kind of work that can run there.
We’re obviously still in this environment–and you have seen us talk about it–where there are certain workloads that just lend themselves to running in certain environments way better than others. I mean that technically and economically. And so our world is one in which one size doesn’t fit all. We really firmly believe we have great X86 offerings, we have great Power-based offerings, and we have great mainframe offerings, and we wouldn’t encourage people to do to a least-common-denominator and just run everything on one system.
The Watson system is a great example of a workload to run on Power. That would not have made a great system to run on z.
Just a second. There was a squinting of eyes. I just want to see what that meant.
Timothy Prickett Morgan: I was just envisioning what size mainframe would be required to run Watson.
Tom Rosamilia: It’s not so much size. The reason for my comment is that the z system has the fastest microprocessor and the fastest uniprocessor performance in the industry. That is not really what we are looking for in the Jeopardy! system. With that system, we wanted to go massively multithreaded, and the fact that we threw 2,880 cores at it and the ability to support four threads per core gave us the ability to chase 10,000 ideas at a time. I wouldn’t run that workload on a mainframe. Mainframes are run for their high I/O rates. This system did no I/O, and it had 15 TB of RAM and the speed between the RAM and the processors is what really gave us the advantage.
End of Part One. Look for Part Two next week.