Power8 Offers Big Blue And IBM i A Clean Slate
November 18, 2013 Timothy Prickett Morgan
Sometime around the middle of next year, IBM will start shipping the initial Power Systems machines based on the Power8 processors. Now is the time to start thinking about the business that Big Blue wants to have with the IBM i platform. And it is also the time for IBM i shops to start thinking about what kind of platform they want from IBM and how they want it to fit into their business and budgets. This is a perfect time to shake things up.
As you will recall, earlier in the summer I went through the details on the 12-core Power8 processor, which should spin at up to 4 GHz and offer about 154,000 units of processing capacity per socket on IBM’s Commercial Workload Performance (CPW) relative benchmark. That estimate is based on early performance statistics for the Power8 chip and assumes that entry and midrange machines will get chips that hum along at that 4 GHz speed. Assuming this is the case, that is considerably more oomph than the 61,500 CPWs available on an eight-core Power7+ chip in the Power 720+ and Power 740+ servers that are targeted at the bulk of the IBM i base.
Suffice it to say, processor performance is not an issue when it comes to IBM i shops. There is so much headroom relative to the modestly sized databases and transaction processing workloads that run in the typical IBM i shop that, in fact, the real problem is excess capacity. There is a huge gap between what IBM can deliver and what most customers require. This is the real problem with the IBM i business, and we all know it. And the question is, how did we get here? And what do we do about it?
Let’s start from the business we want IBM to have, and the buying and user experience we want you to have, and work our way backwards.
As best as I can figure, there are on the order of 150,000 unique customers in the IBM i base. I suspect that many of them are using machines in the Power 520 and Power 550 class, with a certain vintage with Power6 and Power6+ processors being popular in the base. Some shops may even be on older iSeries M15 and M25 boxes (remember those?) and obviously some are on Power7 and Power7+ systems and are not in any mood to do an upgrade unless something really important and useful compels them to.
IBM doesn’t talk about how much money it generates from the IBM i business very much, but I suspect that at any given time, only about a fifth of these customers are actually engaged with IBM or business partners, or about 30,000 shops. I don’t think this is by any means an unusually low percentage of the base to be active.
The typical customer in X86 server land–including Intel‘s own datacenters as well as the clouds run by Amazon Web Services and Rackspace Hosting–keeps a machine running for four years and depreciates it over that time. The difference is that those 4-year-old X86 servers in the fleet are only a quarter of the base, and each year, another quarter comes up for renewal. Yes, in that four years of time Intel can cram something on the order of three to four times more performance into the same box, but for whatever reason, the workloads that run on these machines can consume whatever Intel can make. Or, more precisely, customers think they can even though their in-house software engineers or independent software vendors may not be sure. It is sometimes better to have the threads you might need a year or two from now for a small incremental cost today.
Hope springs eternal.
At IBM i shops, by contrast, IBM has been delivering a big performance boost every three or four years with a modest speed bump in the middle. It is less linear, and the Power5+, Power6+, and Power7+ jumps are not as impressive as the jumps from Power4 to Power5, Power5 to Power6, Power6 to Power7, and Power7 to Power8. These are the jumps that count, and no, I have not forgotten about all the good things that went into Power5+ and Power7+ chips that had nothing to do with boosting CPU performance and that allowed for work to be offloaded from those CPUs and therefore gave them a kind of performance boost.
Believe me, not having to juggle multiple jobs at the same time makes you run faster on one job. . . . Not that this is really an option for most of us in the 21st century. But I digress.
The trouble is that IBM i shops don’t need a lot of capacity, relatively speaking, and they tend to do their upgrades on one primary system, and maybe a backup and development box, every four years or so. So while Intel can count on customers upgrading a quarter of their fleets every year, IBM has a quarter of its customers upgrading one machine every four years. And maybe that has stretched to five or six years in many cases as the amount of raw CPU in a machine goes up and up and up with each generation.
It seems to me that the AS/400 business was very healthy back in the day when the base was larger, workloads were being added to the machine at a steady pace, and improvements in hardware performance were harder to come by.
Before symmetric multiprocessing came along with the AS/400-D80 back in April 1991, if you needed more performance than a base or midrange machine, it cost you because there was no parallelism. You didn’t add extra processors; you had to crank them up and that was expensive to do. But, if you needed that extra 20 percent more performance, you could justify it because the AS/400 was, for all intents and purposes, the data center at most midrange shops. Then along came SMP, and expanding the machine got somewhat easier. And then multicore processors came in 2001, and you could cram two cores into a socket with Power4 and Power5, and IBM skipped right over four cores with the Power6 and soon it was eight cores with the Power7. And here we are up to twelve cores per socket coming down the pike with Power8. That is, in effect, and entire AS/400-S70 machine, circa 2000, inside of a single socket and clocking several times faster on top of that.
I have always said, and I continue to believe, that there is perfect elasticity in the pricing for capacity on AS/400 systems and their follow-ons, and that IBM would be smarter to have a very large base of customers who are not paying top shelf prices for capacity even though they are getting top shelf capacity. It is simple economics, which is particularly brutal in the chip business. Over time, customers will eventually migrate to any cheaper option, thanks to constant budgetary pressure. There is no option.
Intel keeps growing its share of the server business by delivering aggressively priced CPUs that have enough enterprise features to be good enough for most IT shops for most workloads. Amazon Web Services has cut its prices for cloud services 38 times in the past seven years since the cloud was launched, and I learned last week that the company’s Trusted Advisor service sent out over 1 million recommendations in the past year or so to customers showing them how to spend less money on AWS services. Some 700,000 of those recommendations were acted upon, and it is saving AWS customers around $140 million annually on cloud capacity.
Yeah, it sounds crazy, don’t it? But this is the modern era, where customers expect not only more performance, but a price break on computing. This is, of course, what AS/400 shops learned to expect once SMP and multicore processing came to the platform. And this is the game that IBM must play now if it hopes to preserve its Power Systems business.
So here is the deal. With 150,000 customers, there is only so much money available. Say that the average IBM i shop spends $25,000 on a new system or upgrade. Then that is only $3.75 billion, and spread over five years that is only $750 million a year. I do not believe the IBM i business is anywhere near that large anymore, and frankly, I don’t want to think too much about how far it has fallen from its peak of $4.8 billion in processor and operating system sales in 1998. My concern is that there may be 30,000 active customers who are upgrading their machines every four or five years and another 120,000 who are happy to run their very old machines for a very, very long time because they are very reliable and spare parts on the cheap are readily available. That gives an annualized run rate of about $150 million a year for base systems running IBM i if they all upgrade in a staggered fashion, about one fifth a year. My guess is the business is somewhere between these two extremes, and but closer to the lower figure.
Still, 150,000 customers that are loyal to their applications and their platform is nothing to shake a stick at, and there has to be some way to engage all of these customers and get them all to start spending money with the Power8 debut–and keep spending money at a steady rate after that. This is the trick, the real challenge. Adding AIX and Linux to the platform, or low-cost Linux machines or Power IFLs do nothing to directly help IBM i sales, although as I have said, anything that helps the Power Systems line get stronger helps IBM i last longer.
Let’s think about this another way. What if we brought most of these customers to a giant cloud? Imagine there are 150,000 shops out there and that, on average, what they needed was a single core of Power8 to run their current workloads. Assuming that IBM could build a 48-socket Power 870 server, which I think it can, using the 12-core Power8 chip, how many Power 870 machines would it take to cover all of those 150,000 cores? Try 260 systems on for size. OK, so let’s say that the number is really three cores of IBM i capacity are needed on average. Then it is on the order of 780 machines. Call it 1,000 Power 870 servers. That is the size of the IBM i cloud that Big Blue could build to support the entire IBM i ecosystem.
Sounds crazy, right? But here is what I believe. If you get customers on modern machines with modern operating systems, databases, and application development tools–and I don’t care if you call it a cloud or not–and get them on a curve where it gets cheaper as they use capacity, instead of more expensive, IBM can make it up in volume while also cutting prices. AWS is sure doing it for its several hundred thousand customers worldwide. Big Blue has to start thinking more like Jeff Bezos, the founder of Amazon, and less like Thomas Watson, the founder of IBM. And IBM can use its business partner channel to sell it, too.
There are other possibilities, of course. But the important thing is to get customers to spend money and stay in the habit of spending money for a service they love. The trick to not to get less money, but to make IBM i shops feel better about the money they are spending because they get a different kind of service. This, in fact, is what made the AS/400 great so many years ago. It is time to think outside of the box again, and Power8 is the perfect opportunity to do so.