The State Of The IBM i Installed Base, Part 2
February 17, 2020 Timothy Prickett Morgan
In the first half of our analysis of the IBM i installed base, we looked at the number of machines and the number of logical partitions that respondents of the IBM i Marketplace Survey for the 2020 report gave last fall when they took the poll. We did some math and analysis on this to show that there is a large block of customers with lots of machines and lots of partitions that are just as important to Big Blue as those with big, fat NUMA servers.
In the second half of this series about the state of the IBM i base, we are going to get down to brass tacks and look at the vintage of the iron and operating systems in use at IBM midrange shops at least as far as the survey done in late 2019 is concerned. Thanks in part to the encouragement of the late Dan Burger, our executive managing editor for nearly two decades here at IT Jungle, and our assistance in encouraging people to take the survey, IBM i systems software provider HelpSystems has been doing a survey of the IBM i market for the past six years.
The sixth annual IBM i Marketplace Survey report was released by HelpSystems in mid-January, and it was based on a survey that had more than 500 respondents, which is a pretty good number. I participated in the webcast going over the survey results along with people from HelpSystems and IBM, as I have done for the past several years. Tom Huntington, the executive vice president of technical solutions at HelpSystems, hosted the webinar last Thursday, and we were joined by Alison Butterill, IBM i product offering manager, Ian Jarman, who had that job before and is now an executive in the IBM Lab Services division, and Brandon Pederson, who is worldwide IBM i product marketing manager. You can get a copy of the survey report and also listen to a replay of the webcast at this link. We have been interpreting the report results in a series of stories, and we already covered the plans that people have to upgrade their Power Systems hardware and IBM i software in 2020 and of course the system and partition analysis from last week. This week we want to see how the Power9 ramp and IBM i 7.3 and 7.4 ramps look compared to prior generations of hardware and software technology.
Let’s start with the hardware distribution by processor generation from the survey:
In the webinar going over the results, I said that I would have expected that the Power9 ramp would be a little bit stronger given the historical trends for the Power8 iron. That was a misstatement, and I will explain why.
The entry Power9 machines that could run IBM i were launched in February 2018 and the high-end boxes supporting IBM i got Power9 motors in August 2018, so you would have expected that at the end of 2018 when the survey was done for the 2019 report, customers would have some Power9 iron. And indeed, 14 percent of those polled said they had Power9 iron, and at the end of last year for the 2020 report, that share rose to 31 percent of the base. The entry Power8 machines made their debut in April 2014, and the high-end boxes based on Power8 rolled out in October 2014, so we would have expected for the Power8 to make its debut in the IBM i Marketplace 2015 report. And it did. But only 8 percent of customers said they had Power8 iron in the 2015 report, and that only rose to 22 percent in 2016. Now, to be fair, the Power9 machines were in the field three months longer, so that helped the ramp some, but based on this data, the initial Power9 ramp in year one was 1.75X steeper than the initial Power8 ramp, and in year two, the ramp was 1.4X percent steeper.
The Power8 ramp took longer than and built up to the point where close to 60 percent of customers report having at least one Power8 machine, and for the past two surveys (which are really about 2017 and 2018 data), it has been holding steady. This mirrors the long life that the Power7, launched in April 2010, and the Power7+, launched in July 2012, had in the field and as you can see, it is only now that the combined Power7 and Power7+ base is starting to wane, with only 38 percent of customers reporting having these older systems installed.
By the way, no one is trying to suggest that customers have only one machine or only one hardware level installed. The data shows that customers have a mix of iron, and if you add up the percentages, it is a very tight range that is consistent across the six years of the survey at right around an aggregate of 140 percent. The diversity is very consistent as iron comes and goes in the datacenters over the course of five or six years.
The other thing to note here is that machines using Power5, Power5+, Power6, and Power6+ processors, which date from 2004 through 2009, are still represented in the installed base as 2020 was rolling around. Collectively, these machines still were in use at 15 percent of customers.
The one thing we do not know from this data, of course, is the distribution of machines by Power chip make. Survey respondents are only telling you what they have, not how much they have. So, for instance, the response is the same for a shop that has one Power5 machine as one that has 50 machines; both count as one customer with a Power5 machine. We can’t just take the distribution of processors across the more than 500 people who took the survey for the 2020 report, representing their companies, and apply it to the 2,285 machines we think were represented in the survey based on the machine count per customer that we calculated in last week’s story. There is no way to calculate a finer level of granularity out of this data, and there is no way of getting it except by asking each customer to tell about their precise machines and their configurations across the entire company.
Still, this is a good indicator that Power9 had a good year, and it also suggests that, despite the overall downturn we are seeing with the Power Systems business, the IBM i portion is only halfway through its upgrade cycle. So maybe there is reason for optimism on this front even if it looks like some of the “upgrades” coming in 2020 will not be new systems and a switch to Power9 chips, but a move up to Power8 iron or an activation of latent capacity on an existing Power7, Power7+, or Power8 machine or a swap of processing cards perhaps. As we pointed out three weeks ago, our interpretation of the upgrade plans of IBM i shops for the survey done by HelpSystems for the 2020 report depends in large part on what the term “upgrade” means to the customers answering the question. Given all of the data we see, we would not be surprised if the IBM i portion of the Power Systems business does as well in 2020 as it did in 2019, although it probably won’t do as well as it did in 2018. But it is hard to say for sure, given how little data IBM gives out about this business.
What we can say for sure is that IBM has made it a lot easier to get current on iron and stay current thanks to Technology Refreshes without having to move to each release. And from what IBM told us a little more than a year ago, it looks like plenty of customers are skipping every other release and just using Technology Refreshes to get the new features and hardware support as it becomes available.
Now, let’s talk about operating systems. Here is the chart from the 2020 IBM i Marketplace Survey report:
We have to be careful about this operating distribution, too, based on the wording of the question. It asks about the primary operating system level in use at the customer, not for all operating system releases in use at the customer. These are very different questions, and we suspect that some people answered them one way and some the other way. In fact, after looking at all of the past survey reports put out by HelpSystems, I didn’t hear the question clearly myself until just now, and I was reading it as the latter rather than the former. (Sometimes, it is amazing that human beings are communicating at all, and I wonder that we really are with the precision that we think we are.)
What we can see here is that i5/OS 5.4 and older releases have dropped out of the picture, no matter which way you want to answer it, with the 2020 report, and the aggregate of 6.1 or older releases is on the decline. Mysteriously, use of these older releases rose from 2015 to 2016 by a factor of 1.6X and declined in 2017 and held steady for the follow two reports in 2018 and 2019 before falling by nearly half in 2020.
You can see the effect of the skip releasing by some customers in the data. IBM i 7.1 rises and holds to a nice smooth bell curve, IBM i 7.2 is kinda choppy in its adoption and also more muted albeit still very substantial, and IBM i 7.3 is on a smooth curve ramping up nicely. IBM i 7.4 is just getting started, and unless and until IBM makes Db2 Mirror clustering priced for low-end machines, it may have a distribution as the primary operating system similar to IBM i 7.2 over the next few years. We shall see.
What we would really like to see is the distribution of operating systems across machines and logical partitions. As far as we know, PowerVM will not let a machine run more than three different IBM i release levels at any particular release level of its own. But even still, the mix of operating systems could be a lot richer than this chart above implies. And, as I am apt to point out, given that I believe that the people who respond to surveys are also the ones who are most active in the customer base and therefore more likely to be current on hardware and software (as well as read The Four Hundred), this whole dataset could not be representative of anything but what it says: the primary operating system level in use at IBM i shops, not the total distribution of operating system levels in use. They could be very much the same, or not. We can’t know from the way this question was asked. The percents always add up to 100 percent in every year and that is a giveaway that people answered the question the way it was asked, not in the way that would be even more useful.
And it is on me as much as anyone else for not noticing that before now.
We do expect for the overall operating system distribution to be wider than the data for the primary operating system, given that there is a new release every 18 months to 24 months and new Power processors and hardware every 36 months to 42 months and that you can run new releases on older hardware for many, many years. The Technology Refreshes allow the skipping of releases, too, which mitigates against such operating system diversity, so it is hard to figure what forces are working harder on the base here.
All in all, what Big Blue is doing with IBM i is working pretty well, and it sure beats having a slew of shops stuck at V5R3, V5R4, or 6.1.