The annual Red Hat Summit, held in Boston last week, is something of revival tent for open source where the pulpits are plentiful and so are smiling believers. Indeed it’s hard to dispute the powerful innovation springing from open source. The RH Summit, which started 11 years ago as a modest celebration of open source (mostly Red Hat Linux), has since mushroomed into a boisterous, expansive technology ecosystem showcase.
So it’s interesting when a senior IBM exec turns up in a keynote slot. Big Blue’s heritage, at least at the high end, had for years been dominated by proprietary architecture. No longer, said Doug Balog, general manager of IBM Power Systems. The founding of OpenPOWER roughly two years ago, sale of IBM’s x86 business, and the sprint away from the formidable but proprietary Blue Gene (and re-embrace of the battle-tested mainframe) are all part of IBM’s about-face.
Balog is smack in the middle of proving IBM’s commitment to open source and community development. “It’s not an accident I’m here or that IBM is a major sponsor,” he said.
Message to Red Hatters
“The message we’re bringing to this audience, is we are bringing higher value to customers where we can help clients do amazing things. We’re able to do that by supporting an open approach and that’s kind of unique versus some of my competitors in the systems business who are taking commodity infrastructure and swapping the software available and really not bringing in any differentiation. They are simply providing the recipe from Intel in a lot of ways.
“Our value is we have differentiated systems – we have POWER systems, we have storage systems, we have mainframe systems. Those deliver unique capabilities but still leverage open technology to do it. We think that’s the right recipe,” said Balog.
According to the official bio, Balog is “responsible for all facets of the Power Systems’ business including strategy, architecture, operations, technology development and overall financial performance. He is also a current member of IBM’s Performance Team and a recent member of the Strategy Team, which focus respectively on tactical execution and the strategic direction for the IBM Corporation.” Probably doesn’t get much sleep.
Balog’s IBM counterpart is Ken King, general manager, OpenPOWER at IBM. The organization is separate from Doug’s but heavily related because both are based on the POWER8 chip. Both Ken and Doug report up into the Senior Vice President of IBM Systems, Tom Rosamilia.
While at RH Summit, Balog talked with HPCwire on a range of issues including how and why openness informs the new IBM; what the emerging HPC strategy is; how big data and mobility are transforming enterprise computing and why that’s good for IBM’s mainframes (z Systems); and Big Blue’s nascent cloud-based outreach to OpenPOWER developers. He also couldn’t resist taking a few light jabs at Intel and the ARM camp.
Bucking Technology Headwinds
“It really started two years ago with a conversation with Google, Mellanox, NVIDIA and Tyan (a mother board company). Up until that point systems generally ran faster year after year because processor speeds kept advancing,” he said. The group fretted about slowing the rate of performance gains for applications as well as the prospect of brutal commoditization caused by stagnant innovation.
“We said, there’s this great model called open source software. Can’t we take that model and adapt it to open source hardware and incorporate that same community approach to innovation. Yes, at the end of the day IBM will pick pieces from that innovation [for their] products but others will too. It’s going to create choice in the market place and innovation, not commoditization,” said Balog.
Perhaps not a revolutionary idea – they had a model – but giving away proprietary advantage and its associated higher profit margins isn’t easy if you’re not forced to. That’s kind of the point of the free market. Views vary on how successful and how open the new initiative is, but the OpenPOWER Foundation is growing. The gang of five has grown to about 140 members at present according to Balog.
“Jim Whitehurst (Red Hat CEO) and I were discussing his new book (The Open Organization: Igniting Passion and Performance, published in June) and talking about what‘s needed for openness. First, you have to form a [substantial] community because we have seen plenty of attempts at openness where it ended up being a set of family members getting together and nobody else joined. There’s always more to do but I feel good about the community we have built so far.”
“The next phase is getting a community to bring innovation to the space and we are starting to see that. At the [first] OpenPOWER summit in San Jose [in March] there were 15 companies who brought POWER-based motherboards, all very different, all very innovative, all really targeted at the cloud or HPC companies,” said Balog (see IBM’s First OpenPOWER Server Targets HPC Workloads, HPCwire).
Show Me The Money
“Now we have to start to move in the monetization direction. How do you take this community, take this innovation, and start to transform it into products and sales. How do [OpenPOWER] members who want to build on top find that balance between innovation and openness and turn it into monetization. We are all publically traded companies so at the end of the day we want to see some sales.”
IBM helped its OpenPOWER cause recently by luring longtime industry executive Sumit Gupta away from GPU pioneer (and founding OpenPOWER member) NVIDIA. Gupta is VP, HPC & OpenPOWER and reports to King. He had been GM of GPU Accelerated Data Center Computing for NVIDIA.
“He was a perfect fit to come in and lead our HPC OpenPOWER business. He’s well connected in the industry and has a great business mind so that’s the role he has. He’s the guy Ken King and I look to and say OK you’ve got two big wins where’s the next five or ten or whatever the number is and where do we go from here.”
It doesn’t hurt that Gupta has extensive knowledge of accelerators which play a critical role in IBM Power Systems plans for HPC (see What IBM says about Gupta’s role, HPCwire).
“So [moving away from Blue Gene] was one of the strategy changes made two years ago. Historically we had built these wonderful engineering marvels called Blue Gene systems, beautiful, well-engineered machines but they are monstrous in size. Quite custom you would agree. In the best cases we shipped a few systems and didn’t lose money; in most cases we lost money. It’s a tough market.”
“We said why don’t we have an HPC strategy that takes our standard one and two-socket systems and buck those up with accelerators through CAPI attached (POWER8 Coherent Accelerator Processor Interface). CAPI plays a big role here, or NVIDIA attached with NVLink. I think accelerators are a big opportunity here and that’s not just IBM hardware but Altera and Xilinx and NVIDI, etc.
“That’s the kind of HPC system we’ve targeted so it’s much more about a rack full of scale out systems with accelerators. Obviously you need the code optimizations, so you’ve got to pick your target industries where they are willing to take their code and start heading down a path of leveraging accelerators. Some have seen that light already and some haven’t seen the light yet but I think more and more are,” Balog said.
One benefactor of all the openness changes is the mainframe. It’s always been around, but also absorbed its share of condescension over the years. Porting Linux to IBM’s z-Systems line is opening doors to new uses. Container technology, Hadoop, and a myriad of other ‘open’ technologies are becoming accessible on mainframes.
“Think about the evolution and transformation of IBM, not as a company though we could talk about that, but also from a systems business and how this aspect of openness is really transcending the way platforms like the mainframe continues to drive growth, [especially in large enterprise environments.]
“[The mainframe] has been around for 50 plus years. Obviously, it’s very different today than it was 50 years ago although you still run the same aps from 50 years ago and that’s one of the miraculous aspects of the mainframe platform is the commitment to architecture continuity while enabling new work. One of the biggest drivers in addition to Linux [on the mainframe] is mobile transactions for z-Systems [such as in financial services.] I mean who actually goes to a bank these days,” said Balog.
IBM’s key HPC targets are unsurprising – government, oil & gas, financial services and increasingly life sciences. Balog emphasizes it’s a much more economical way to go, to take what’s already in the portfolio and bring in accelerators.
It will be interesting to watch how IBM fares in the TOP500 list in coming years. IBM had 153 systems (roughly 30 percent) in the T0P500 List last November including four in the top ten. IBM does have a couple of big HPC wins recently, one a DOE and the other with Science & Technology Facilities Council (STFC) in the U.K.
“We’re not really that focused on being in the TOP500. It’s not that we are shying away from those opportunities, but it’s not about the number of the score anymore. A lot of our focus is on the sweet spot for POWER and the marriage of HPC in those industries where it really is about the data analytics. That’s where the POWER architecture shines through even with this addition of the accelerator model. As you can see from the accelerator [possibilities] our approach is quite different than Intel’s in terms of it’s an open approach. All are welcome to bring their best acceleration technology. We didn’t see the need to go spend $16B dollars,” said Balog.
OpenPOWER, of course, isn’t the only processor-based ecosystem out there. Intel remains the giant everyone aims for. To say it has been less than spectacularly successful is just plain wrong. On the other hand, the market does seem hungry or at least open to more choice. Stir in the slowing of chip advances (i.e. the much discussed demise of Moore’s Law) and growing worry over power consumption and the suddenly potential opportunity for non-Intel contenders seems more realistic.
The ARM camp is one contender. It’s been a huge winner in the mobile device space where ARM’s reduced power requirements are critical. Traction in the server market has been slow, but release of a 64-bit design (ARMv8) is making matters more interesting. Just a week ago, the Mont-Blanc Project at the Barcelona Supercomputer Center (BSC) fired up a prototype running on ARM suggesting it is possible to get high performance from the architecture. Mont-Blanc is exploring more energy efficient approach towards achieving exascale computing.
ARM Needs a Body of Support
“We keep asking ourselves about it. Our view of ARM is it had promise, we were watching it, but I think we’ve seen a lot of the ARM server companies start to fold up tent and move away from it. Part of it is a weak core design, if I could call it that, I don’t mean that disparagingly, it’s just that that’s what it is. That’s why it is in all of our mobile devices.
“It hasn’t built the server ecosystem that clients might want to look at. So we just haven’t seen it mature at the pace it might have been able to mature and it’s been around for a while. You know AMD recently sort of declared they are moving away from it,” said Balog.
HP’s Moonshot server line has a model with ARM which has a few wins, a notable recent one at the University of Utah. Part of the challenge is to interest the developer community. Balog asked wryly about Moonshot, “But has HP ever delivered much of the ARM stuff really. They quickly touted that and went right back to Moonshot is a bunch of Intel servers.”
“Could ARM and POWER partner? I don’t know. We continue to ask ourselves that question. We’ll see. It’s low power. Do we think about power issues in the OpenPOWER community and what’s down the line, sure. [But so far] we aren’t seeing power consumption as a major issue to deployment. It is the balance between do you take a slightly stronger core that can run oodles of performance benefits over an ARM or an Intel processor and therefore it’s got a littler more energy consumption or do you go with a lot of systems with a really weak core.
“We continue to watch the space. We’re all for openness so I think they help chip a way at Intel at the low end and we chip away at middle- and high-end. The market wants choice and that’s sort of the fundamental thing we hear from the cloud companies,” said Balog.
Clearly a big challenge, and directly related to IBM’s presence at the Red Hat Summit, is engaging the open source developer community. They need to be convinced of IBM’s commitment and to be able to play on the POWER architecture. To some extent, IBM’s efforts there remain nascent, but growing.
“[Developer outreach] is more and more cloud based, no surprise, by providing access to POWER infrastructure in the cloud. [The idea] is to leverage benefits of a cloud versus [forcing] everybody to have their server on their desktop. A couple of weeks ago in China and we launched an open developer Linux platform in the cloud with accelerators called SuperVessel and it’s come one come all (See, IBM Introduces SuperVessel, HPCwire).
“You can do development, try out accelerators, try out POWER, try Linux on power, and it’s available free. We will expand it to the rest of the world over time. We have some things in the POWER development platform today, we have some slight differences but as POWER goes into software which it will here very soon, there will be Linux on POWER that will be another opportunity for us to provide free connection to developers.”