How To Join The Power Linux Evolution
November 2, 2015 Timothy Prickett Morgan
It is hard to remember that IBM was not exactly sitting on the sidelines when Linux swept over the datacenter in the early 2000s in the wake of the dot-com bust. Big Blue saw the rise of Linux early on, among its supercomputer customers, and it was unsure how to preserve its revenue streams from AIX and OS/400 systems while at the same time embracing Linux. Here we are 15 years later, and it looks like IBM finally has its Linux act together on Power.
The question we all have is this: Will it make a difference? Hope springs eternal, so we are hopeful that the efforts of IBM and its OpenPower partners will provide a Power-based platform that is competitive with the Intel Xeon platform that is the default setting in the datacenters of the world for most modern software. The latest Power Systems LC machines, made by Tyan and Wistron with some input from IBM and its hyperscale and HPC customers, go a long way toward closing the gap. The systems software stack, including the baseboard management controller and BIOS for the systems, has been reworked with the help of Google and gives the machines the look and feel of an X86 system. The performance of these machines is better than or equal to that of the Xeon machines they compete with (and all vendors put their thumbs a little on the scales when they do such comparisons, we realize), and the pricing is within spitting distance, too.
The problem is–and this is something that the long history of IBM and many other IT players, some of which are no longer with us, some are just skeletons of what they were–it takes more than technology to get customers excited. They have to feel part of a movement, part of a trend. Part of the future.
IBM has countless great technologies that were better than open alternatives–Systems Network Architecture made the TCP/IP stack look like an unsophisticated toy in the datacenter, but it had the virtue of being open. Token Ring had more bandwidth and lower latency than Ethernet, but markets like open and choice. The X86 processor architecture was never open, and it is not an “industry standard” as Intel likes to say, which implies the industry got together and endorsed an open thing, but rather a “de facto standard” by virtue of its ubiquity. ARM is similarly not an open standard, but is at the least licensable and malleable, which is why there are a half dozen vendors jockeying for a position in the datacenter with their ARM variants.
To its credit, IBM is trying very hard to build a future for the Power Systems platform, one that is more open and inclusive, even if it may seem a bit alien to IBM i shops. IBM has been shipping Linux partitions on machines using OS/400 since 1999, and in fact, the first virtualization that allowed Linux to run side-by-side on a Power-based server was created by IBM’s Rochester Labs and the hypervisor that allowed this was derived from OS/400, which already had logical partitioning for OS/400 workloads. While Linux has seen some action at OS/400 and IBM i shops, the most successful Linux installations (if you can call them that) are tools that run in the Portable Application Solutions Environment, or PASE, an AIX runtime embedded in OS/400 and IBM i. Whether companies know it or not, IBM’s own TCP/IP and Java stacks inside of IBM i are actually AIX versions running in PASE, completely transparently, and MySQL, Node.js, PHP, Ruby, and a number of other tools popular on Linux platforms run in PASE within IBM i.
This is good, but for certain kinds of workloads–particularly data analytics jobs such as Hadoop, Spark, Redis, and so forth–what customers will need to do if they want to go with Power platforms is learn to use Linux as it is and put it on Power Systems LC iron.
I know what you are thinking. Why would anyone do that, when nobody is going to get fired for buying a rack or two of Xeon servers from Hewlett-Packard, Dell, or Lenovo? Well, here is the thing. You, IBM i shops, and your brother and sister AIX shops, are the tip of the wedge. And if you cannot be convinced to adopt Power platforms for these new workloads, think about how hard it will be to convince the companies that have been buying fleets of X86 iron for one or two decades? AIX and IBM i shops adopting Linux0-based Power Systems LC iron (or their equivalents from third parties) is a kind of enlightened self-interest. The supercomputing base at the high end, and supported by the governments in the United States and Europe, are not enough to guarantee the long-term viability of the Power platform. The 100,000 AIX shops and the 125,000 IBM i shops in the world have to be the vanguard.
Last weekend, I was talking to a successful IBM i reseller who has branched out into archiving and vaulting services, partnerships to sell ERP stacks, and providing a slew of other services, who is growing his business nicely and is profitable, and he wanted to know what was in it for him with these Power Systems LC machines. At first, I figured it wasn’t much. But then I got to thinking. All of the enterprise customers in manufacturing, retailing, distribution, education, and government who are using IBM i platforms for their transaction processing are going to be wrestling with mountains of unstructured data that will never end up in their transaction processing systems and yet is useful for running the business and doing things such as recommendation engines, search, analytics for ecommerce and other applications that companies park out on the web.
Selling a rack of Power Systems LC iron for these modern workloads to 225,000 customers would be a lot of iron. Let’s take the “Habanero” Power S812LC two-socket machine made by Tyan as an example, since it is aimed at workloads that need lots of local storage and to offer a two-to-one node advantage over Intel Xeons of equal performance. A configured Habanero server with two 10-core Power8 processors running at 2.9 GHz with 256 GB of memory and two 1 TB disks lists for $17,000 and costs $14,000 on the street including a Red Hat Enterprise Linux license. (A two-socket Xeon E5 v3 server with two 10-core chips has a street price of $11,600, but it does less work and offers slightly worse bang for the buck based on SPECint_rate tests.)
For fun, let’s assume all of these IBM i and AIX shops want to do data analytics on lots of unstructured data and would normally put this on Xeon servers these days. Across those 225,000 potential Power Systems LC shops, and let’s say they average a half rack of machine each, which is a mere 10 systems. That works out to 2,250,000 machines, or about a quarter’s worth of X86 server volumes, and that also works out to $31.5 billion in revenues. Cut 40 percent off this for deep discounts, and it is still $18.9 billion. Spread it out over five years, and it is still $3.8 billion a year. Cut it in half to be pessimistic–and then cut it in half again to be downright grumpy–and it is still a potential $1 billion a year in revenues. Add in some storage and then licenses for commercial-grade Hadoop, Spark, Redis, and such, and it might be a tidy $2 billion business on shipments of 112,500 machines a year. These would be shipment numbers that IBM has not seen since the late 1990s with the Power platform, by the way. Just to give you some perspective. And as that unstructured data kept growing and more use cases drove more workloads, it would grow even as Moore’s Law crammed more compute and storage in a box every three years or so.
This scenario sure beats a sharp stick in the eye (or rather, in the IBM i), and this is money that IBM i and AIX resellers can–and I would argue should–learn how to chase.
Going after data analytics workloads with these Power Systems LC machines would not be an easy task, mind you, but IBM indeed wants to peddle these new-fangled data analytics and NoSQL wares to its Power Systems faithful. It has just not figured out how to talk about it properly. Frankly, the company is trying so hard to woo Google and other hyperscalers with Power, which might have a big advantage when it comes to running search indexing software so I can understand IBM’s focus. But IBM can’t forget its own customers. It can’t keep making the same mistake it has made through the decades, which is always chasing the big customer and forgetting about the smaller and midrange ones. It is better and safer to have 225,000 happy customers than 100,000 annoyed ones looking for the exit. In fact, it would have been better to have 500,000 happy customers with Power Systems, but don’t get me started about how IBM won the Unix battle but lost the systems war in the data center. . . .
Like I said, hope springs eternal. One need only look at Apple, which has faltered twice, to see that.
Now that IBM has cheap Power Systems iron that is competitive with X86 iron, it needs to make AS/400-like appliances out of these machines with Hadoop, Spark, Redis, and such operational software and integrate them tightly with the transaction processing systems based on IBM i and DB2 for i or AIX and DB2 or Oracle. IBM has to make data analytics easy and transparent for its own base, and it needs its partners to understand how this works and how to sell it and it needs to teach its customers why they need this. And while I commend IBM for having a Power Systems Linux proof of concept system deal–see announcement letter ZWAM5618B for details–in the United States and Canada, this is not enough. IBM has to be dramatic. Like saying the first 1,000 customers who have IBM i and AIX as their primary transaction processing platforms and who agree to use BigInsights Hadoop and Spark or Redis or what have you on Power-Linux iron get the machines at 50 percent off.
Pump up the volume. Create a wave. Create some excitement. Show some real growth in Power Systems, despite currency headwinds. If IBM can spend $2 billion to buy The Weather Company to fluff up its Watson analytics business, how much should it invest in expanding the presence of Power platforms at their most natural customers, IBM i and AIX shops?
It is a tall order, and I said as much to that IBM i business partner. But it is something that makes as much sense as just giving up the market to X86 platform makers. The long-term success of the Power platform might depend in large part on how well or poorly Big Blue gets IBM i shops and AIX shops on board with Power-Linux machines. This is important stuff, and no one knows better than the AS/400-IBM i community about the importance of applications in driving budgets.
IBM could, of course, just do all of this on its SoftLayer cloud. But again, Big Blue has not put a stake in the ground and said something like 25 percent of all servers going into SoftLayer will be Power-based this year and 50 percent next year, and 75 percent the year after that, and 100 percent in four years with the long-term goal of unplugging every X86 server SoftLayer has over the long haul because Power is better than Xeon.
If you want the market to believe, IBM, you have to believe, too.