The Power Systems Base Is A Little Less Rusty

February 27, 2023 Timothy Prickett Morgan

Another three years, and another upgrade cycle that can bring about the modernizing of the Power Systems iron that supports the IBM i customer base. Or, more precisely, another four distinct upgrade cycles that customers are on as they move off legacy iron to something more current than what they have.

What we mean by this is that some customers are stuck on older releases because they are cheapskates by necessity or they have applications that can only run on older releases like OS/400 V5R3, i5/OS 5.4, or IBM i 6.1. Others are stuck on IBM i 7.1. Or IBM i 7.2 or IBM i 7.3. So there is not just the current upgrade cycle from any prior generation of hardware to a shiny new Power10 machine, but also an upgrade cycles from Power6 to Power7 or Power7+, and another upgrade cycle from Power6 or Power7 or Power7+ to Power8, plus an upgrade cycle from Power6, Power7, Power7+, or Power8 to Power9.

You can see these upgrade cycles all superimposed on the same dataset over the past nine years from the IBM i Marketplace Survey, which Fortra (formerly known as HelpSystems and we will stop reminding you of that soon) puts out every fall and publishes in January of the following year. But as you well know by now, we don’t think that the data that comes from the survey is perfectly representative of the entire IBM i-Power Systems base. We think that the 120,000 customers running IBM i on Power Systems break into two camps, with a quarter of them, or 30,000 unique customers, being what we call active customers and the remaining 90,000 unique customers being laggards.

This split is admittedly a hunch, but it is one based on anecdotal evidence from customers and business partners who have as long as we have been following the base for three and a half decades have always indicated to us that there was far more vintage System/38, System/36, AS/400, iSeries, and System i iron in the field than any survey data ever showed. We think that surveys such as the one performed by Forta (as well as others) show the customers who tend to stay relatively current with hardware and software, who mostly are not caught in the La Brea tar pits of vintage and unsupported software from third party vendors who may not even exist anymore or who charge too much money to upgrade applications to current releases. We think that such customers are the ones who read publications like The Four Hundred, go to COMMON and other user group meetings, and take the time to do surveys. And we think that the Forta survey, to be specific, is absolutely representative of what they are doing in terms of Power Systems and IBM i release levels.

We know that the laggards are quite a bit further back, although by how much is not clear. To try to figure that out, we took the dataset of Power Systems distribution by Power CPU family and did a little witchcraft on it to figure out what the machine base at the remaining 90,000 sites might look like.

First of all, the way that HelpSystems and now Fortra asks the question, companies are allowed pick whatever Power CPU families they have installed, so the percentages always add up to more than 100 percent. Usually it is around 140 percent, which implies and average of 1.4 machines across the customers who take the poll. In the 2022 survey that was published in January 2023, the total added up to 129 percent, which implies an average of 1.29 machines across the base. There are many, many customers who have one machine, many who have two or three for high availability and disaster recovery, and some who have more than three up to dozens. We did some math on the machine count distribution and came up with a model that reckons the number of machines in the field per customers is actually somewhere around 2.7X, with some very large shops with lots of machines pulling up the class average a lot higher. We think some people are answering for all of their machines and others are answering for their primary machine. We don’t know precisely how to square this disparity.

But what we did reckon is that assuming that the laggards are running about four years behind on hardware, we can take the survey data from four years ago and apply it to the share of machines at the 90,000 laggard sites, use the data from the current survey to figure out where the active portion of the base is at, and then add these together to get a proper distribution of Power Systems iron running IBM i across the world. We also built a model to show the machines that are collapsed into the “Power6+ and older” category so we can see Power6/Power6+, Power5/Power5+, and Power4/Power4+ transitions over time. In older surveys, these machines were labeled separately.

This mathematical witchcraft is an important exercise because the original dataset coming from the nine annual IBM i Marketplace Survey reports would give you an absolutely wrong – and very much rosier – view of the Power Systems-IBM i installed base than we think is real.

But, before we get into the math, let’s take a look at the distribution of machines in the 2023 IBM i Marketplace Survey report:

It would be great if 6 percent of the installed base of Power Systems machines running IBM i were already running Power10 processors, but just on the face of it, we don’t believe that is true anymore than we believe that 6 percent of the installed base of X86 servers in the datacenters across the world are running Intel’s latest “Sapphire Rapids” Xeon SP CPUs, which started shipping unofficially to the hyperscalers and cloud builders in volume last summer and which were only formally launched to the rest of the world in January.

We similarly do not believe for one second that an incredible 67 percent of the customers polled (representing a smaller somewhat smaller portion of the base of machines in the field because you need to divide that by 1.29X) after nearly four years of sales are based on Power9 processors. You don’t believe that, either, and if you are on a Power Systems machine with a Power8 and earlier processor, you are in good company. We assure you of that. And double ditto for Power7 and Power7+ machines, which we think are all over the base.

When you take all of the data in all nine surveys and break out the older machines based on Power4 through Power6+ processors distinctly, here is what the raw survey data looks like:

In this raw data, Power7 and Power7+ peaked in 2016 and has been sliding more or less steadily down, Power8 peaked in 2019, and we think Power9 will probably peak in 2023. The earlier machines – Power6+ and older – peaked a long time ago, but HelpSystems was not doing surveys then. We can guess, but it is a lot of work for not much value since that was ancient history.

We don’t think the laggards are anything like this, and we have to make some guesses about what they were doing back in the early years of the surveys. But as you can see, the data is pretty regular, and there is no reason to believe the pattern has deviated too far.

The laggards move slowly, and they move up one or two hardware generations, not four or five, for obvious reasons as we pointed out above. And when you do a time shift backwards for the laggards and add this to the active portion of the base, you get a historical distribution that looks like this:

That is a much different picture, isn’t it? Power7/Power7+ actually peaked in 2019 across this complete installed base that we simulated, not in 2016, and Power8 may only be peaking now or perhaps in the 2024 report. And Power9 is still on the rise and is running four years behind the Power8 ramp and still only comprises a little more than 20 percent of installed machines — a lot less than Power8 machines and still less than Power7/Power7+ machines. Older machines are still in the base, but have come down fairly linearly, and in the base of Power5/Power5+ machines have come down very fast since 2014.

What this chart says to us is that IBM should worry less about what IBM i versions and releases it can support on PowerVM logical partitions on a particular Power10, Power9, and Power8 system and more about how it can use the Technology Independent Machine Interface, or TIMI, to make a Power10 processor look like a Power9, Power8, Power7+, or Power7 processor, and maybe older stuff. Rather than trying to update everyone to more modern hardware and operating systems, why not convince older releases of OS/400, i5/OS, and IBM i that they are actually running on whatever hardware they expect? It mostly means hiding more modern features anyway.

Why it took us so many decades to think of the problem this way, and a solution to it that is different, is beyond our ken. But we just did. Why can’t TIMI and microcode make any processor look like any other processor? With such an approach, you could in fact move everyone to Power10 and just leave their software the hell alone. All of it.

This Issue Sponsored By

Table of Contents

Content archive

Recent Posts

Subscribe

Pages

Search