The State Of The IBM Base 2022, Part Three: The Rusting Iron
March 28, 2022 Timothy Prickett Morgan
In the past several months, we have been drilling into the results of the annual IBM i Marketplace Survey that HelpSystems does every fall and then reports on each January. We have been taking our time going through the results, and in a number of cases we have been doing our own spreadsheet magic on top of the raw data to provide what we think is better information that describes the current state of the IBM i base.
In our first story, we talked about the distribution of operating systems over time, spanning from the 2015 report to the 2022 report, and rather than just regurgitate the raw data and cogitate about it, we re-examined this data with a supposition we have been making for many years now. To whit: We think that there are about 30,000 truly active IBM i customers in the world, by which they stay reasonably current on hardware and software, they pay for maintenance, and they read newsletters like The Four Hundred and they participate in surveys.
The rest of the user base, which numbered around 150,000 unique sites in 2012 and which has been around 120,000 sites for the past several years (and does not appear to be shrinking substantially any more), is what we call IBM i laggards. They use older tools, they run on older systems and they employ older IBM i releases – and sometimes even OS/400 and i5/OS releases, if you can believe it.
We drilled down into the security issues that IBM i shops are facing, the evolving use of programming languages, and even calculated the average age and experience levels of those attending the 2022 IBM i Marketplace Survey webinar in January and applied that data to the pool of IBM i people in the world, which includes programmers, system admins, managers, and system architects and which we think numbers around 384,000 people worldwide, to see what kind of crisis we might be facing in the coming years. Separately, we have talked to compensation experts to see what the prevailing wages are for IBM i personnel, and talked about the disparity of compensation based on platforms, experience, and geography.
This week, we take on the installed base of AS/400, iSeries, System i, and IBM i machines that in the world. We know from IBM roughly how many unique customers there are, and from the HelpSystems report we can see what the distribution of machine counts are across the customers who took the survey. Using our active-laggard split, and the distribution of machines in the HelpSystems poll reports from 2015 through 2022, and making some assumptions about what the laggard base might look like – fewer machines, older machines – and forward casting some finer-grained data that used to be gathered about older systems (in the 2015, 2016, and 2017 surveys), we have come up with what we think accurately reflects the current state of the worldwide installed base of machines running IBM i and its predecessors and how it has changed over time since 2015.
Just like the adjusted operating system distribution estimates that we did showed that three quarters of the base is on relatively old to downright creaky operating system releases – meaning IBM i 7.1 and earlier – this analysis shows that there is still a large base of ridiculously vintage equipment running out there. We ran this data past some business partners who service customers with all kinds of machines, and who don’t focus solely on selling old equipment but it giving customers the iron they need for their applications, and they said that this data is consistent with their experience. And that the data for the active part of the base also matches exactly what they see among their customers. This is anecdotal evidence, but this is the only kind that we have aside from raw survey data that we do not think accurately reflects everyone’s reality in the IBM i base.
Before I dive in, a few words. By calling sites laggards, there is no judgment on my part. It is a statement of fact, and as I tell my children when they need to hear something they might feel is judgment, I tell them: “This is data, not judgment.” Many of you are laggards, and I am too in certain spheres. I would still be running Windows 95 if Microsoft would have let me, and I moved to Windows 7 begrudgingly and to Windows 10 begrudgingly and I bought a new deskside machine in January and I just don’t want to deal with a move to a new machine and Windows 11. But I will get to it when I have a moment.
I also know full well that The Four Hundred has always served the laggards as much as the actives, and that is also intentionally so. We try to present a mix of data and tech tips that are useful for both camps, and anyone that is spanning them. (All distinctions are, to a certain extent, arbitrary.)
But more than anything else, we think it is important to see the base as it really is, and not as it appears to be from such survey data, and as I have been saying all year, see this as an opportunity for business for all of us to help get customers on newer machines with newer and more secure software stacks while preserving the applications and databases that companies have spent decades creating that run their businesses. We want to encourage everyone to move into the future where it is appropriate, and we have ideas about this, but before we can do that, we have to try to get a dataset that we hope better reflects reality. We have to deal with what is before we can get to what can be.
Let’s start with the raw data from the IBM i Marketplace Survey. Here is the 2022 report’s machine distribution:
And here is that finer-grained older data from the 2017 report:
That second chart is very good at showing how each generation is growing or shrinking over time, and the first one is better for seeing some overall trends, but we think a line chart that brings all of this data together and that estimates how the base of Power4/Power4+, Power5/Power5+, and Power6/Power6+ machines have changed over time between 2018 and 2022 is important. So we looked at the rates of change and broke apart the “Power6 or Older” – it is really “Power6+ or Older” if you want to be precise, and we do – into these finer grained datasets. And here is what the trend data looks like:
This all looks fine, right? The older generations of Power iron – Power4 through Power6+ – have declined to nearly nothing, Power7 and Power7+ are in steady decline, matching the decline of the earlier machines but four years out of phase of Power6/Power6+, and Power8 ramped as expected, and Power9 ramped almost exactly at the same pace. This looks healthy. Comforting. But something is wrong.
If you take the piece of the survey data that relates to the number of machines that respondents say they have installed – 1, 2 to 5, 6 to 10, 11 to 30, 31 to 60, and 61 or more – and take the midpoint of those ranges and then apply that distribution over the 140,000 unique customers we estimated were running OS/400 and IBM i operating systems in 2015, you end up with just under 740,000 machines. If you do the same thing for the 2022 data, you get around 455,000 machines.
There is no way there are that many OS/400 and IBM i machines in the world. We have not seen numbers that high since the 1990s for the installed base. So something is out of whack. When you apply the distributions that HelpSystems saw in its surveys to the 30,000 active sites, and then use a more modest distribution of machine counts per site for the laggards – 50 percent have one machine, 20 percent have two machines, 15 percent have three machines, 10 percent have four machines, and 5 percent have five machines – you have 110,000 laggard sites that have just under 300,000 machines in 2015 and 90,000 laggard sites that have around 250,000 machines in 2022.
Now here is the tricky bit. If you apply the distributions of machines by Power CPU family across time to the active sites as shown in the raw data collected by HelpSystems, and then assume that the laggards are back in time – we reckon that the 2018 distribution of Power processors in the actives is a good proxy for the Power CPU distribution in the laggards today, but we may even be a little generous there – and then add it all up, here is what we think the installed base has looked like between 2015 and 2022:
First of all, it makes a dolphin.
Second, we would be able to predict where the laggards might go using the active site CPU distribution for 2019 and 2022 and applying it to 2023 through 2027 – if it were not for the fact that we think certain customers hit a wall at IBM i 6.1 and can’t move and other customers – and I mean a lot of customers – have hit an IBM i 7.1 wall and can’t move ahead, either. In a sense, these two parts of the base are like the System/36 base from days gone by.
Obviously, this “dolphin” distribution is a radically different pattern of machine changes in the base. The Power7 family kept growing through 2019, and we presume this would be caused by companies moving from Power5 family and the Power6 family needed to move to new iron, but they had to hang back on iron because their software was not able to run on more modern releases. This would also match the large base of IBM i 7.1 users we projected in our operating system analysis, and the IBM i 6.1 base would be running on those remaining Power6 family and Power5 family machines. (The family means the base CPU and its “+” kicker when IBM shipped one.) This data shows that the Power7 family base is now the same size as the Power8 base and the growing Power9 base is the same size as the declining Power5 and Power6 families. I didn’t seek to make these things true, it just happened in the process of doing the active and laggard split and making assumptions – reasonable ones, I think – about what that CPU distribution looks like in the base of machines running at laggard sites.
I don’t like the way this looks any more than you probably do.
Perhaps these two different charts might represent a kind of bookend for the base – a worst case and a best case scenario. But based on the stories I have heard from many people, I believe that chart two reflects the reality in the IBM i base overall.
The question now is: What do we do about this? I have some thoughts, but I will share them on another day.