Some Thoughts About IBM’s Next Generation Platform
April 9, 2012 Timothy Prickett Morgan
The software people at IBM are now firmly in charge of systems development at the company and are apparently trying to bring iterative software development techniques to hardware. The idea, in short, is to make hardware more modular as well as the systems software stacks that run upon that hardware so IBM can integrate new functionality into systems in an iterative fashion, perhaps several times a year, rather than a big bang system change every two or three years. This is not the dumbest thing I have ever heard.
By picking up the pace, IBM could accomplish a number of things. For one, the company can get innovations into the field faster. And for another, it will smooth out IBM’s sales and make the revenue stream from its Systems and Technology Group more linear instead of the sine wave–maybe it is a cosine wave?–it has been on for decades. The latter is an interesting idea, but I am not sure if the market will consume hardware this way. And third, this iterative approach with a focus on the software harkens back to the early days of the System/360 mainframe in the late 1960s and the launch of the midrange AS/400s in the late 1980s when Big Blue clearly understood that it was not just bending metal, but creating complete, integrated systems that were tuned to work well together.
Back in February, I walked you through the idea of “refactoring” hardware designs, as IBM’s Ambuj Goyal, general manager of development and manufacturing for the Systems and Technology Group, explained it to IBMers sometime last year. I also explained some of the aspects of the Flex Platform concept machine–a kind of hybrid blade/rack box with a more modular hardware design than current machines. We now know that IBM is getting ready to launch something that is called the “Next Generation Platform” internally and that is being called an “expert integrated system” on April 11, as I told you in last week’s issue of The Four Hundred. It is not clear how the Flex concept machine and the NGP system are related at this point. They could be very different things.
Just for fun, I want to riff on what little I know about NGP system and provide some thoughts and hopes for what may–or may not–actually end up being in the product.
First of all, I have not forgotten that the name of the System x division was actually changed to Modular Systems back in a previous March 2008 reorganization. No one uses that name, of course, this being IBM. That was the same time that IBM formally and finally merged the System i business with the System p business to create what we called the Power Systems division for two years. In July 2010, when Software Group ate Systems and Technology Group to create a single Systems and Software Group under general manager Steve Mills, Power Systems and mainframes were united under Tom Rosamilia. (And if I were a betting man, I would guess he is in the running to succeed new IBM CEO Ginni Rometty someday.)
Here’s a thought. This Next Generation Platform should be created and controlled by the Modular Systems division, one that is separate from the politics of the other quasi-divisions. Do you honestly believe that by putting one executive in charge of two divisions a merger actually happens? I think such a thing takes years, particularly with such different cultures. If the Next Generation Platform is IBM’s future, then maybe IBM should make a new division, create a new culture, and source components from the other IBM divisions as it needs them.
Here’s another observation: If the rumors are true, the Next Generation Platform will look an awful lot like the “California” Unified Computing System from Cisco Systems. That’s not necessarily a bad thing, mind you. The good news for both IBM and Cisco is that the networking giant has already been doing the hard sell on bladish-rackish servers backed by converged networking for servers and storage with integrated management for the past three years, and 10,000 customers have bought into the idea and are generating more than $1 billion in product revenues a year for Cisco at the current run rate. The revenue stream could be twice that as Cisco exits calendar 2012, since the revenue run rate doubled from the end of 2010 through the end of 2011. IBM’s Next Generation Platform will likely be very different from the UCS platform, but the concept is not new and that will help Big Blue sell.
Some Free Advice On The Next Generation Platform
One of the great things about being a newsletter editor is that you can offer all kinds of free advice, like other armchair quarterbacks and generals. (One of the unusual and fun things about my job is that occasionally I actually get paid, in real cash money, for my free advice.) And as you might expect, I have some free advice for Big Blue as it contemplates a possible server future based on the Next Generation Platform.
Make sure the SmartCloud uses the exact same iron. If IBM is designing a modular system that has X86, Power, and possibly System z processors and is espousing the capital and operational benefits of such a platform, then it should be the backbone of the SmartCloud. In fact, if I were running IBM, I would be deploying Next Generation Platforms in the SmartCloud first and then ask customers to help shakedown those systems before rolling them out commercially to the data centers of the world. Giving SmartCloud processing capacity for half price or even free to NGP testers might be smart before rolling the product out to commercial clients. (There may not be time for this, of course, depending on when the MGP systems are shipped.)
I would even go so far as to say that maybe IBM should only support NGP iron in the SmartCloud. The whole point is for IBM to somehow leapfrog Amazon‘s Web Services compute and storage clouds, and if NGP can’t be better than Amazon’s mix of homegrown servers and machines made by Silicon Graphics, then there is not much of a point to playing the game. And speaking of which, just like Amazon has funky data center designs, IBM needs to have Next Generation Data Center designs that marry the new platforms with new–and modular–data centers. You don’t put new wine in old bottles, after all. I don’t think you need to use shipping containers; those form factors are great when you need to take a data center on the road to a remote area, but otherwise they are not the right dimensions for either the humans or the server and storage racks.
Make sure the IBM i operating system is supported on the Next Generation Platform–and on SmartCloud, too. This seems pretty obvious, but Big Blue sometimes forgets the obvious.
I’ll make this simple: Anything commercial–meaning, not a supercomputer or a workstation–that runs AIX on Power should be running IBM i on Power. Period. Whether it is a system in your data center or cloudy capacity served up from one of IBM’s data centers.
Make sure there is a tower version of the Next Generation Platform. Do you know why blade servers are a niche product still? Because they were never designed for an office environment, like tower servers were. It was stupid to put tower servers on their sides and pile them up in data center racks, and hence we got rack servers in the late 1990s in the data centers (at least for X86 iron, since the original IBM AS/400s and 9370s were rack-mounted way ahead of that time). It is equally stupid to take a big wonking blade server chassis, turn it on its side, put wheels on it, and then put sound muffling foam on it to make it an “office blade server.”
What IBM needs to do is make a tower server that has the same modular concept as the NGP system but shrinks it down to one box that still fits under the desk, beside the desk, or in the closet. Instead of using a two-socket, half-width motherboard, maybe the basis of this machine is a much smaller microserver board with a single socket, and you can slide one, two, or three of these into the tower chassis. You use the cheaper low-end chips in this box and allow customers to upgrade to faster single-socket variants. Maybe you even put an optional slot in the board for a chipset module that can plug in and convert two one-socket servers into a two-socket server that can share memory. IBM owns the Power architecture and it can do this. If not, buy ScaleMP and do it with software.
The idea is to not abandon the small and medium businesses that are the backbone of the old AS/400 business and the now much smaller Power Systems-IBM i business. Embrace SMBs and give them the same benefits of modular hardware and integrated networking and virtualization. Hewlett-Packard and Dell are sure not doing this, and Oracle, like Sun Microsystems before it was eaten by Oracle, does not believe in tower servers. IBM has an opportunity to re-engage with SMBs in a way that it arguably has not done since the AS/400 launch in 1988. And no, the SmartCube appliances, which debuted in late 2008, don’t count all that much.
Make sure that the Next Generation Platform is aggressively priced to take market share. Despite all of the operational and integration benefits, blade servers are perceived as being too expensive relative to rack servers. And that is why blades servers only account for about one fifth of server revenues these days.
If IBM wants to displace HP and Dell in the volume server space and also prop up its AIX, IBM i, and z/OS businesses, it has to do so with a better product at a lower price. This is how IBM went from a joke in the Unix server business in the late 1990s to the dominant position here in 2012. IBM offered much better bang for the buck with its AIX on Power machinery than HP’s HP-UX on either PA-RISC or Itanium platforms or Sun Microsystems’ Solaris and Sparc combo, and the Power4, Power5, and Power6 machines also usually offered more processing, memory, and I/O scalability, too, than the HP and Sun alternatives. That’s not to say that HP and Sun did not field good machines, but IBM was aggressive with both technology and price on the AIX side.
I would add that IBM was in a unique position to milk the OS/400 side of its business and absolutely overcharged for green-screen processing capacity to help subsidize the capturing of Unix market share from HP and Sun. In short, you paid for IBM’s success, readers of The Four Hundred. And as best as I can figure, from 1998 through 2012, more than half of the OS/400 customer base has moved to other platforms. In a meeting somewhere inside of IBM Somers, I would have argued that it was critical not only to maintain those 275,000 customers that were using AS/400 iron back in 1998, at the peak of the market, but to focus on increasing IBM’s share of the X86 and Unix portion of the processing at those OS/400 shops. IBM could have made the same revenue and market share gains by treating these customers better, but I doubt it could have made the same profits. Hence, IBM did what it did. And now, something like 155,000 former OS/400 shops count HP, Dell, and others as their primary platform providers.
So, what do you think about all this? Click that contact button and speak your mind.