Plotting Out A Power Systems Resurgence
September 8, 2014 Timothy Prickett Morgan
IBM is on the brink of shedding its System x server business and freeing itself of the inherent contradiction of wanting to push its own systems and wanting to also be a player in Windows and Linux X86 server arenas that represent the vast majority of shipments and a dominant slice of the revenue pie in the systems market. Soon, Big Blue will be solely a peddler of Power Systems and System z machinery.
Does IBM really believe that it can take on the X86 platform with the maintenance of the mainframe base and a steady and continuing expansion of Power iron the likes of which we have not seen since the dot-com boom?
It is hard to say what IBM believes, but clearly this would be a very good thing for IBM in general and for its IBM i and AIX customers and partners in particular, and it is certainly something to be desired if for no other reason than to give Intel some competition. The stakes are much larger than this. The future of both the IBM i and AIX platforms is not in question–IBM will continue to support both platforms on Power8 iron for many years, and Power9 chips are being designed now to extend those platforms into the future. But the kind of future these two platforms have is dependent, as always, on the revenues and profits that the overall Power Systems business can generate and IBMâ€™s ability to plow those profits back into the business to expand sales and marketing and add differentiating features to both Power hardware and the IBM i and AIX operating systems.
Ironically, or perhaps expectedly, the future of the Power Systems platform may depend upon the success or failure of Linux on Power. This is probably not a situation that makes most IBM i shops comfortable, given that IBM started moving Linux to Power and mainframe systems back in 1999 and most OS/400 and then IBM i shops have pretty much been allergic to Linux on Power. If these diehard IBM shops are not willing to fire up more than a partition on their Power Systems machines and use Linux to run infrastructure workloads or application servers–or better still, whole applications written in PHP or Java or some other language–then why would we reasonably expect for shops that would otherwise buy X86 machines to run Linux on Power?
IBM has done a great deal of work over the years in conjunction with the Linux community and with Red Hat and SUSE Linux, the two biggest vendors of commercially supported Linux distributions, in particular to get Linux running well on Power machinery. Generally speaking, Linux offers the same or better scalability and performance as AIX on a given system, even the largest Power 770, 780 and 795 machines with hundreds of cores and over a thousand threads. Java and database performance of Linux on Power is akin to that of AIX, too. This is better than can be said of IBM i, which IBM certainly could make scale across larger numbers of cores and threads but has not in recent IBM i releases, presumably because there are no longer a sufficiently large number of customers hitting the performance ceiling to justify the work. In the late 1990s and early 2000s when AIX and OS/400 scalability were on par and some very large customers had multiple of the largest OS/400 machines to support their databases and applications, IBM had to keep OS/400 expanding in terms of core, thread, and NUMA socket scalability.
I have said it before and I will say it again. IBM overcharged OS/400 shops for their processing capacity and software licenses so it could undercut Sun Microsystems Solaris and Hewlett-Packard HP-UX platforms, but in doing so, it lost about half of the OS/400 customer base. Some of them moved to AIX, but most have moved to Windows, often on IBM iron but probably as many times as not on HP and Dell iron. This might have been a good strategy for the executives who ran the Power Systems business in the short term, as IBM ate market share like crazy and both Sun and HP were unable to compete with IBM on price.
But IBM could have given AIX customers and OS/400 customers similarly great deals and kept the OS/400 base growing while also eating Unix market share and cultivating Linux on Power for further expansion. IBM is addicted to share buybacks and continual growth in net income and refuses to create a strategy where it can make it up in volume like its competitors. That is how Intel and Microsoft do it, after all. And what it comes down to is the fact that International Business Machines never figured out how to be a low-cost manufacturer of anything. Dell and HP are not great at it, either, so IBM’s issues here were less apparent until Supermicro, Foxconn, Quanta, and others showed what they could do in terms of lowering the cost of making a system and eating volumes.
I could go on and on. If IBM had opened up the Power architecture back in 1991 with the PowerPC initiative, if it had open sourced Systems Network Architecture and Token Ring back in the 1980s, which was incredibly better than the TCP/IP and Ethernet stack of the time, OS/400 could have been a flagship operating system. Imagine if OS/400 had been open sourced before Unix even got a toe-hold in the late 1980s.There are alternate universes to explore there, for sure. In our spare time, over beers, perhaps someday we can do this.
It is with these thoughts in mind that I am overcome with a pesky kind of annoyance that the future of the Power Systems platform and the IBM i operating system are linked to the success or failure of Linux on Power, but to be honest, I don’t have a better set of plays for Big Blue given its competition and the assets it can bring to bear.
We want continued investment in Power machinery, but IBM can’t really afford the vanity of having its own foundries anymore and, moreover, its Power8 chips are wedded to IBM’s own chip processes, which use its own copper, silicon-on-insulator, and high-k metal gate technologies. If IBM had kept control of the game console chip business and not lost this to Advanced Micro Devices, if it had ramped up Power Systems volumes by competing aggressively with both OS/400 and AIX machines and adding in Linux to the mix, and if it had not lost the Apple chip business to Intel, then things might be different. If the U.S. government was still spending hundreds of millions of dollars a year on massive supercomputing projects and if IBM was still pushing Power-based systems for these petascaleclass workloads and striving to reach exascale, that would have helped, too. But somewhere along the way, maybe four or five years ago in the belly of the Great Recession, IBM lost its hardware heart even though it is still capable of creating machines of great elegance, sophistication, and scale. Maybe this is what happens when software people and services people take over IBM.
Now, IBM would be wise to be working with a third party foundry–perhaps GlobalFoundries or Taiwan Semiconductor Manufacturing Corp–to get Power8+ and Power9 chips moved over to their most advanced processes. The 16 nanometer process at TSMC, which will be used to etch Oracle’s 32-core Sparc M7 processors next year as well as a handful of different ARM server chips, looks promising. If no one will buy IBM Microelectronics, then it should make all of the Power8 chips if will ever need and then shut it down and write it off rather than continuing to lose money on it.
The fact that there are Power processors being designed for future Power Systems is much more important than the fact that IBM does the wafer baking. Indeed, it is more important than ever that Big Blue pick up the pace of innovation to not only stay on Intel’s heels, but to somehow leapfrog Intel in a big way as it did with Sun and HP starting with the Power4 generation in the Unix wars at the tail end of the dot-com boom and the beginning of the bust. Instead, IBM is off doing science projects like the Synapse neural chip or Watson cognitive systems. These are all well and good and may represent the future of computing for certain domains. But in the meantime, companies are still buying clusters to run real applications to run their businesses.
The Scale-Out Power8 machines launched in April have a lot of good things going for them. Because IBM changed the byte order of the processors to match that of X86 processors, it is easier to port X86-Linux workloads to Linux on Power. IBM has created its own variants of the KVM hypervisor and the OpenStack cloud controller to make Power-based clusters look and smell like X86 iron, and it has added the Ubuntu Server variant of Linux from Canonical to the Power8 machines. As far as I can tell, IBM is pricing the Power-Linux variants of the Power8 systems aggressively enough to compete against two-socket Xeon E5 machines, but I have this bad feeling that IBM will be tempted to charge a premium for the forthcoming Power8 Enterprise machines that are expected before the end of the year. That would mean larger AIX and IBM i shops would be footing the bill for Linux expansion on Power, and that is a net losing strategy. IBM has to keep those AIX and IBM i customers happy, too, so they will remain in the fold.
I have nothing against Linux, and in fact I think it is a remarkable thing that there is finally am operating system that spans most server processors and is available on most systems, regardless of architecture or maker. This has never happened before in the history of the computing business. IBM could offer freebie Linux partitions on IBM i machines–and set them up with RHEL, SLES, or Ubuntu Server so they are ready to rock the moment the machine is fired up. This would get customers started in using Linux alongside IBM i from the get-go. IBM could even go so far as to give customers incentives to move key groupware and database workloads from Windows Server on X86 to Linux partitions on Power. Just like IBM used to have massive porting efforts to get Solaris and HP-UX workloads moved onto AIX, it could get clones of Exchange Server and SQL Server up on Linux partitions and help customers move. IBM would no doubt want to charge a lot for such services, but it would be better by far to think of this as a long-term investment, just as IBM did with Linux workloads running on mainframes.
Here is why. It is now 15 years after Linux first appeared as an experiment running on logical partitions inside of the VM operating system on System/390 mainframes at Marist College, and Linux now accounts for about 8 million MIPS of aggregate processing capacity out of a total installed base of 30 million MIPS in the entire global System z mainframe base. This didn’t happen by accident. Activating a core of Linux is about a fifth the price of firing up a core for z/OS, and a Linux license from Red Hat or SUSE Linux, while being crazy expensive compared to the same software for X86 machines is considerably less expensive than z/OS.
I think the real lesson here is that Linux can help save the Power platform by getting it to grow again, but it surely cannot do it alone. IBM is going to have to get aggressive about promoting IBM i and AIX as well if the platform is to do more than just sink with the rest of Unix.
I have some ideas on how this might be done, and if you have any, hit that Contact button and let me know your thoughts. We are all in this together.