Rebuilding The Bottom Of The Pyramid
March 25, 2019 Timothy Prickett Morgan
In last week’s issue of The Four Hundred, we told you about how Big Blue had extended the life of the Power8-based entry Power S812 Mini, announced on Valentine’s Day last year specifically to give entry IBM i shops a cheaper alternative than buying the Power S814 or Power S824. It seems to me that IBM needs to do some rejiggering of the way it bundles and prices this entry machine to get the installed base of customers using vintage hardware and operating systems to get current and stay there.
We are under the impression that the number of machines running IBM i 6.1 and earlier releases – including the venerable OS/400 V5R3 and i5/OS 5.4 releases that have not been supported by IBM for years – is larger than many want to admit. Admittedly, for the customers who buy IBM or third-party support for their midrange systems, the presence of these older releases is pretty small. And if you believe that, then there is no real problem to solve.
But based on the limited data we have seen, only about a fifth of the base is on maintenance, so we really have no idea what the other four fifths are doing. You cannot just apply the ratios of the part of the base you know to the much larger part of the base you don’t know. That may show you a probable baseline share of the overall base that might be using vintage systems, but this is a floor, not a ceiling. That vintage base could be running on much older hardware and operating systems than we think. The people who take polls and read newsletters are the ones who are, generally speaking, more active in the community and more engaged with vendors. So by definition we know more about them. We all know there are people who don’t do any of these things, and it is not because they don’t love the AS/400 and their progeny. It is because they are often from shops with only a few people or one person and they have a lot to do just to maintain the applications and systems that they have.
We are doing some straw polling among ISVs and resellers and will do our own poll to try to get a better sense of this. Let’s assume for the sake of argument that this vintage base on older releases and iron it is larger than polls have suggested. What do we do to get these small IBM midrange shops on modern wares? We can start with keeping the Power S812 Mini alive longer and tweaking it to get these customers moving ahead. Once they are on a modern machine, we can work to start consolidating Windows Server and Linux workloads running on X86 iron onto Power Systems machines.
A single core Power S812 Mini, which we detailed here, has just under 10,000 CPWs of performance, which plenty enough to get started. But with a 64 GB main memory cap (which is also imposed on the four-core Power S814 system), which IBM most definitely imposes to encourage customers who have more pressing memory needs to move up to a two-socket Power S824, this is not enough memory to accommodate some big workloads we want to add in the future. So we want IBM to take off the memory cap, but for now only for customers who buy the updated Power S812 Mini under a special migration deal we are proposing. Later, we can talk about doing a microcode or firmware update that unlocks the memory on all such entry Power systems that have had governors put on them.
The next thing we need to change is the way IBM i and Linux are packaged and priced on this Power S812 Mini. One of the selling points as part of the upgrade will be that the system has four cores on it, and three out of the four of them will be available for incremental IBM i licenses later or Linux. Under this deal that I am proposing, at least one of the cores has to be configured with Linux, and specifically, Red Hat Enterprise Linux, which Big Blue will own by the end of the year if all goes well with its $34 billion acquisition.
For these entry machines, IBM i is sold in the P05 tier on one core, and it costs $2,245 with 90 days of Software Maintenance added in. If you back out that Software Maintenance, it costs $1,996 per for the P05 license for that core. If you add in three years of extended maintenance with 24×7 coverage, it costs $6,047 to cover the operating systems, a basic software stack of compilers, database tools, and hardware support as we outlined last year in talking about the relative price and performance of Power8 entry systems. If you load up the Power S812 machine with everything you need, then it costs $41,042. That included software access for 25 users (five bundled into the base license and 20 added separately.) Add another $3,567 and that covers Software Maintenance for another two years, bringing the total system cost to $44,609 and that covers costs over a timespan that we can reasonably expect for these supposed vintage OS/400, i5/OS, and IBM i shops to keep this new machine. This works out to $743 per month for the whole shebang. That’s a little more expensive than buying the 2019 Challenger R/T Scat Pack with the 392 horsepower, 5.7 liter SRT V8 Hemi engine, but nowhere near the $71,695 Challenger SRT Hellcat Redeye, which has a 797 horsepower in its 6.2 liter V8 Hemi engine.
The entry price should come down to give these vintage customers some money back to help port their applications, which is going to be the real problem. And moreover, since we want them to grow their workloads, it can just suddenly rocket into the P10 and P20 IBM i software tiers where they go from spending tens of thousands of dollars to hundreds of thousands of dollars to support workloads.
So, here is what I propose. IBM sells the Power S812 Mini hardware in a base configuration with one core on a four core, 3 GHz Power8 processor fired up. Add 64 GB with only two memory slots populated (leaving six others open to be used in the future), plus a reasonable 10 Gb/sec Ethernet card (come on, IBM, 1 Gb/sec is pretty crappy these days), six 500 GB disk drives, and two flash drives. Yeah, I know the Power S812 Mini didn’t have flash drives. But this one will, and it will do so while only costing $15,000 for the hardware. Over five years, that works out to $250 a month for the hardware. If you add more memory or different disks or whatever to this base configuration, you pay the differential cost – and no artificially boosting prices – at the same prices that IBM’s Linux on Power customers pay with the Power Systems LC iron.
Now, we need to rejigger IBM i pricing. First we are not going to charge per core for systems software. And second, we are not going to distinguish between a base IBM i license not including the database and a Red Hat Enterprise Linux license, and we are not going to count users or cores any more, either. Forget all of that. Licensing per core hasn’t made sense since there was logical partitioning, and it certainly doesn’t make sense in a system that can have ten logical partitions per core. We just want to count sockets, and we want customers to deploy whatever operating system they want. This proposed special Power S812 Mini machine always has to have at least one IBM i partition and one RHEL partition, and they can be any size and number the customer wants up to the limits of the iron, spanning across the four cores that are actually in the system. Take that governor off too. We are not counting users to have a 5250 tax, just like Linux doesn’t count users.
So the question becomes, what is that price for the IBM i and RHEL subscription? I can make this real easy. It costs $1,299 per year for a RHEL license on a server with up to two-sockets, plus another $240 for extended update support. That’s $129 per month. The IBM i compiler and database tool stack costs roughly another $6,000, which works out to an even $100 per month. So that is it, for any machine with one or two sockets. So what is the integrated Db2 for i database worth? For the purposes of this deal, it doesn’t matter because we are bundling it in for free as an incentive to get customers modern. But let’s say the database is worth twice as much as the operating system, tool, and compiler stack. Call it $250 per month for DB2 for i to get a nice round number, plus $125 per month for the systems software, plus $250 a month for the hardware. That’s $625 per month, and as customers upgrade to newer Power9 machines once they need more oomph, this licensing follows them. No P10 or P20 or P30 tiers that zap the budget. If they stay under the two-socket barrier, the price is the same each month for the software. And if you have four sockets someday, you just double the two-socket price, just like Red Hat does with RHEL licenses.
The point is that this pricing is totally neutral to the technology in the socket. Power8, Power9, Power10 – I don’t care. Get the best hardware you can to run the software. What better incentive to stay current on both hardware and software can there be? Perpetual licenses are so 1980s, and they only worked when customers were upgrading hardware frequently and rapidly changing the software at the same time to take advantage of advances in the software (such as database enhancements or the addition of Java back in the day), which justifies both the hardware and software upgrade fees on an already installed system or the cost of shiny new licenses on completely new machines. People want and expect monthly subscriptions on premises and hourly subscriptions on the public cloud. And paying $625 a month for a two-socket or smaller IBM i system is pretty reasonable.
We can’t let this all loose on the IBM i market all at once, of course. This deal is only for those who have IBM i 6.1 and earlier releases of the operating system and only for those who have Power7+ and earlier hardware. Let’s be optimistic and assume the IBM i customer base has 150,000 unique customers (as some at IBM have suggested) and maybe there are something like 200,000 machines in the base. Let’s also be pessimistic and say that at least half of these machines are vintage by these definitions above.
At $625 per month for 100,000 machines, if you could upgrade all of them to Power S812 Minis, then that would be $750 million a year, and on a five year contract, that would work out to $3.75 billion. If each and every one of those customers had an HA backup to the IBM i instances on the IBM Cloud, that might be another $1 billion over five years, and if they had other compute and storage services, it might be another $1 billion. This would be an annuity-like revenue stream, and it would be spent by customers who are always looking for ways not to invest in IBM i platforms but as stuck with them at the very least or extremely loyal to them at the best but don’t see the need to change. And it is hard to quantify just how much money might be made as these customers moved up into the midrange and high end as they expand their workloads, pulling stuff off X86 iron.
It can be done. It should be done. And at these prices, there is a good chance that it might be done.