The IBM i Base Is Ready To Move On Up
January 22, 2018 Timothy Prickett Morgan
Even the most eager hyperscaler or cloud builder in the world has to let a server sit in place, running workloads, for at least three years to get good value on the dollar spent, and in many cases, machines are repurposed to be in the fleet for longer than that. And no company can accept the risk of changing all machines at the same time, for obvious reasons other than cost, because such a change could be very disruptive to any business. Enterprises tend to have less spare iron and fewer techies, and so they are even more conservative, and IBM i shops, who generally do not mess with success, are even less inclined to make changes for the sake of change.
This is certainly borne out in the 2018 IBM i Marketplace Survey, which has been put together by HelpSystems for the past four years and which has become one of the key ways we get some insight into what IBM i shops are thinking and doing. There are plenty of IBM i shops on vintage Power5, Power5+, Power6, and Power6+ iron and on older i5/OS and IBM operating system releases, but there are companies that have stayed more current. With the economy going relatively strong, the Power Systems business on the upswing, and Power9 systems looming large, there is no better time for Big Blue to get the marketing and sales programs in place to help the base of customers still on vintage systems move ahead.
Let’s talk about the vintage iron first. According to the 650 sites polled by HelpSystems for the survey, you can see from the chart below that there is plenty of old iron still out there in the base:
IBM contends that the average customer upgrades their machines every three to five years, but the distribution in the installed base – even from this dataset that HelpSystems has gathered, which we think tends to be the more active customers, not vintage customers who, for myriad reasons are on older hardware and software – shows it to be much older. There may be many companies that upgrade their iron every three to five years, but there are plenty that are not doing it. The Power6 machines are more than a decade old, and the Power5 machines are pushing 13 years in the field. That is a long time for any system to still be working, and it is a testament to the fine engineering of IBM Rochester that they still are. Even the Power7 systems are now pushing 7 years.
I don’t think this is a good idea, mind you, and heaven knows that I kept my servers in the field for this long when I had to pay for them myself. So I understand.
If I had to guess, it looks like the base of Power5 systems (which includes Power5+) and Power6 systems (which includes Power6+) is moving on up to relatively inexpensive Power7 and Power7+ iron that is widely available as other companies have moved up to Power8 iron in the past four years. (There are no Power8+ machines that can run IBM i or AIX. Those are Linux-only machines and really aimed at hybrid CPU-GPU nodes for supercomputing and machine learning workloads.) The trend data that HelpSystems put together for the past three surveys certainly bears this out; keep in mind, shops can and do have more than one machine, so there can be overlap within a shop where there is a mix of vintages blended together, perhaps with newer iron being used for production workloads and older iron being used for backup systems in a high availability cluster or for developers to create and test code.
What seems obvious from this data is that the change in the share of customers with Power7 and Power8 iron has not changed much in the last year, and I think that this is because companies are waiting to see what the Power9 machines have to offer. Plenty of customers expected for IBM to get Power9 machines into the field that could support IBM i and AIX as well as the big endian version of Linux by late 2017, but the only machine that made it out the door in December was the “Newell” (sometimes code-named “Witherspoon”) AC922 supercomputer node. IBM is pretty vague about what the rollout of Power9 machines supporting IBM i will be, but did give us a head’s up last summer that we should not expect such iron until early 2018. By then, it was worth the wait for lots of shops to do just that, and it looks like they did. The revenue figures for the prior three quarters sure reflect this, with Power Systems revenues on the decline even as Linux-only systems have been on the rise.
The good thing about servers is that they can be a few years old and still support newer operating systems, and just like for other operating systems, IBM i shops generally run more modern releases than the vintage of their hardware would suggest. It would be so much easier to correlate this if IBM literally named the operating system releases for their initial hardware that they supported. (There is a sort of rough correlation.) This is what I think is a perfectly normal distribution of operating systems that would also be reflected in any survey done of enterprise customers using commercial Linux or Windows Server operating systems. I don’t think a survey would find a huge number of laggards – they are by definition not as active and are less likely to take polls or read newsletters, and in some cases they may be unaware of the platform they have under their applications – and the HelpSystems surveys don’t find the laggards in the Power Systems base that are running OS/400 V5R3 or i5/OS V5R4 to the extent that we know, from anecdotal evidence, they exist in the actual installed base of 120,000 shops using IBM i or one of its predecessors on some kind of Power iron.
To put it plain – and I said this on the webcasts I did with Alison Butterill, product offering manager for IBM i, and Ian Jarman, business unit executive for Power Systems Lab Services, who have both run this business for many years along with Steve Will, chief architect for IBM i – I think the base of ancient as well as vintage operating systems is much larger. This represents the distribution, perhaps, for the top 30,000 or 40,000 sites, but the bottom 80,000 or 90,000 sites probably skew more to the earlier releases. We just hear about customers on V5R3 and V5R4 too many times.
This is an opportunity as much as it is a problem. Assuming these customers on old iron and software can be found and are receptive to moving ahead – and that is two big assumptions I realize – then there should be a concerted effort to get them current and then keep them current. Regression and compatibility testing for application software is a big, big deal and I do not mean to diminish this huge obstacle in any way. So, someone needs to create a real IBM i cloud with lots of capacity and the kind of redundancy and high availability that is required at a price that makes sense. If IBM is willing to sell a Power9 server with four or eight GPUs on the cheap (like dimes on the dollar) to the U.S. government for Oak Ridge National Laboratory and Lawrence Livermore National Laboratory because that is an investment of sorts that made Power9 possible, then a real, large, scalable, and reasonably priced IBM i cloud could be used to get those 80,000 or 90,000 customers current. Money they did not have to shell out for hardware acquisitions and systems software licenses and support could be used for code modernization and migration instead. Then, after they made the move, they might decide they want their own machines and move only HA backup to the cloud. Selling a few thousand $25,000 or $50,000 servers a quarter is small potatoes. Upgrading 80,000 or 90,000 customers all to modern iron and systems software would be impressive. A feat. And would boost not only Power9, but the Power Systems business as well as the IBM i platform.
It is time. In fact, it is long past that time. If IBM can make machines that compete on price with X86 iron and beat it on performance, then it can use the same economics to either build a worldwide IBM i cloud itself or have someone else do it. How much can 450 million CPWs of aggregate capacity really cost, which would do the job I think. IBM sells the hardware on Power S814s at around $2.00 per CPW, and IBM’s own costs have to be much, much lower than that. Call it $1 billion at list price for the iron, just to be conservative, including systems software. But remember, you don’t have to build it all at once, and you get paid to build it. Say it makes $300 million or $400 million in the first year and only costs $200 million a year to build. It is a win-win for everyone. As it grows, it is quicker to build the next bit of capacity until the job is done.
It is either this way to get companies modern or give away the systems software – which costs nothing – and slash the hardware to cost and try to get companies to commit to moving ahead, a task that will cost tens of thousands to hundreds of thousands to millions of dollars for most IBM i shops on top of that system cost. This is why people can’t move ahead. But we have to try, or lose them to Windows Server or Linux some day. This opportunity of vintage systems is also a danger.
The IBM i Base Not As Jumpy As It Has Been (2017 survey)
The Feeds And Speeds Of The IBM i Base (2017 survey)
IBM i Trends, Concerns, And Observations (2017 survey)
Finding IBM i: A Game Of 40 Questions (2016 survey)
IBM i Marketplace Survey Fills In The Blanks (2015 survey)