A Power Systems Wish List For The Year Ahead
January 18, 2017 Timothy Prickett Morgan
It is the beginning of a new year and that means a bunch of different things. First, we are not quite as rested as we would like, but we are also ready to get on with it. It usually means that people are making resolutions to change their own behavior for the better and making prognostications about what may or may not happen in the future.
Sometimes we are not in a mood to prepare or predict, and as this year gets rolling I am in much more of a mood to prescribe. Sometimes, you can’t wait for the future to happen all by its lonesome, and you have to help it out with some effort and energy.
To that end, we ask you this: What do you want from the Power Systems division, and IBM more generally, in 2017? Hit that contact button above and let us know. In the meantime, I will tell you some of the things that I think might be useful in the coming year to help bolster the IBM i platform and to make all of our lives easier.
With this being a big year for the Power platform in general and the Power9 chip and the OpenPower consortium specifically, we are excited about the prospects of a revitalized and growing Power ecosystem, which is a big benefit to the IBM i community. Now I am going to get out my soapbox, so get comfortable.
At the moment, there are two distinct kinds of Power platforms. The first is the historical one that supports the old microcode and PowerVM hypervisor that was necessary to run OS/400 and its IBM i follow-on as well as all of the variations of the AIX Unix variant and big endian Linux distributions from Red Hat and SUSE Linux. Starting with the Power8 chips almost four years ago, IBM shifted to a new set of microcode, called OPAL, and the OpenKVM variant of the open source KVM hypervisor. These machines, known as the Power Systems LC line, can only run little endian Linuxes from Red Hat, SUSE Linux, and Canonical. While the plain vanilla Power Systems machines and the new-fangled Power Systems LC machines have the same underlying hardware, they are not allowed to run the same software stacks.
None of us are stupid. IBM has some very sound economic reasons for wanting to bifurcate the Power Systems line. Legacy IBM i and AIX customers pay a premium for compute, memory, and sometimes storage compared to the buyers of those Linux-only machines. The lower-priced Power Systems LC machines are designed to compete against Intel Xeon servers running Linux, and they offer similar or better bang for the buck, at least the way that Big Blue does the math, at the system level with hardware, software, and support added in. The Xeon platform still has an advantage when it comes to compute density and lower power draw and thermals, but the Power Systems LC platform has a lot more memory bandwidth and a lot more threads that it can bring to bear to squeeze more performance out of each core in the box.
To one way of thinking, IBM is getting a little more revenue from those IBM i and AIX customers who are not getting such generous pricing, which is good for someone trying to get a bonus at the company. But we think that the good that comes from this incremental revenue is more than offset by the extra costs of supporting two distinct Power Systems lines and further alienating IBM i and AIX shops who know they are being gouged compared to Linux-only shops.
We have said this before, and we will say it again: IBM should port AIX and IBM i to the new OPAL microcode and OpenKVM hypervisor and make them true peers to Linux, and allow IBM i and AIX shops to better argue to keep their platforms. Getting an incremental 30 percent more revenue from 1,000 customers is not as good over the span of five to 10 years as getting half as much revenue from 30,000 customers who spend every year because investing in the Power platform is easy and affordable. IBM has somewhere between 125,000 and 150,000 IBM i shops, so it says, and I think the real goal should be to convert each and every one.
This brings me to my next wish. IBM should be pushing very aggressively to modernize the hardware in every IBM i and AIX account, and I say AIX because these customers are part of the revenue stream that IBM i shops benefit from. There has to be a way to make an entry Power9 box that is affordable and that is appropriate for each of the IBM i and AIX shops that have not updated their hardware in years. These installed bases need to be made current and kept current. There is some investment here, but a Big Blue obsessed by cognitive computing, data analytics, high performance computing, and other aspects of the IT sector is not thinking about telling customers it already has about the bright future they have with the Power platform because it can unite so many disparate workloads. The “i” in IBM i is for integrated system, after all. And we are suggesting that IBM remember this and provide a truly integrated and single Power Systems platform that can run IBM i, AIX, or Linux in either endian mode. If IBM i can run System/36 and System/38 RPG and COBOL code as well as relational databases from the dawn of time, there must be a way to reunite and reintegrate the platform.
I know what you are thinking. This will take big investments from IBM to accomplish. But we would counter argue that if IBM can get IBM i and AIX running on any servers that come from OpenPower platforms, the Power ecosystem will be stronger.
Maybe the answer is to move all but the biggest IBM i shops to the cloud on such reintegrated Power Systems machines. If Google consumes 50,000 Power9 machines a year, it will be the largest consumer of Power9 systems since retailer and financial services giant Sears was back in the late 1980s and early 1990s, and it stands to reason that its Google Cloud Platform should be able to run IBM i and AIX instances as well as Linux instances. Why not?
Many service providers have built their own clouds, and they would no doubt also benefit from a reintegrated Power Systems lineup. The volume economics and the ability to pick multiple suppliers for hardware has been a key to the hegemony of the Xeon processor in the datacenter. The Xeon chip has nearly 90 percent revenue share and over 99 percent shipment share of the worldwide server market. Isn’t it time that IBM actually see how this happened, and emulate that? IBM is not making its own Power Systems LC machines, and I don’t think manufacturing is as important as selling and supporting, although I like it better when companies make their own stuff.
Those public clouds, like Google and Rackspace Hosting, who are building their own Power9 systems will only support IBM i and AIX if it runs on the new OPAL microcode and OpenKVM hypervisor, so that is reason enough as far as we are concerned to put the money and effort into porting IBM i and AIX to this new system software stack. We need a single Power Systems platform again, and a single platform that runs on premises or on any cloud and that affords application portability across it all.
If I had one wish, that would be it, and then if I had two it would be to convert the installed base to it as quickly as possible.
And yes, I am aware that the Enterprise-class systems are different by nature. But maybe that is something that needs to change, too. It looks like Intel is consolidating down to one Xeon platform with “Skylake,” and maybe IBM should have done the same thing with Power9.