January 29, 2014

IBM’s New Deal Captures Refocused HPC Strategy

Nicole Hemsoth

It’s been quite a month of news for IBM, with the sale of its x86 business to Lenovo, followed by some intense questioning about what this means for their vision (and future) with HPC customers in academia and government in particular. And while some might call this the centerpiece of a shift in strategy, something bigger is looming on Big Blue’s horizon, and the processor piece is only one sliver of the story arc.

But Dave Turek, IBM’s Vice President of Advanced Computing says that ultimately, and yes, even after the Lenovo news, their business will march along to the HPC drum, but with some new beats added to an old tune. After all, as he reminds, IBM was never the only x86 vendor supplying servers to the government and universities. Further, these commodity approaches might not be as well equipped for a data-defined future–one that Turek says requires a honed sense of overall workflow instead of mere flops. Accordingly, IBM is unfurling a grander approach to big science (and big business) problems that blends subtler hues into the HPC server portrait–leading to what might be a completely different picture in the years to come.

Specifically, IBM will be melding some (yet unnamed) upcoming technologies with their vision of data-driven systems that make the concept of workflow and end function of the user requirement paramount. The challenge there isn’t going to be about technology as much as merging these concepts and moving “classic HPC” from its traditional focus. For IBM, the shift involves a wide-ranging view of the entire data lifecycle and, not surprisingly, significant investment in Power.

He noted that IBM’s proposition going forward is “to attack the entire workflow from the perspective of how data is acquired, managed, governed and analyzed in many different areas of the HPC infrastructure, not just the server. The consequence of that is that the nature of what servers look like in the future might change a bit.”

In other words, what IBM sees going forward looks a lot like verbiage around Watson and their Smarter Planet array of technologies—parallels that Turek made explicit during a conversation this week in the wake of some news about a new partnership with Texas A&M that meshes data analytics with high performance computing via IBM-hosted (cloud-delivered) Blue Gene/Q power managed with Platform Computing and leveraging GPFS.  This provides a way for the company to show off the blend of all of its priority tools says Turek, from the file system, the Platform software, and the ability to drive data on a range of applications. The collaboration is, according to IBM, aimed at “improving extraction of Earth-based energy resources, facilitating the smart energy grid, accelerating materials development, improving disease identification and tracking in animals, and fostering better understanding and monitoring of our global food supplies.” Again, all of which have elements aligned with IBM’s Smarter Planet initiative.

“All the servers are on the periphery and the data is in the center of the proposition,” explained Turek. We wanted Texas A&M’s strategy to mirror IBM’s strategy towards data-centric high performance computing that creates an infrastructure that lets them bring different architectures to address the problems at hand. This deal is characteristic of one of the shifts that’s happening in IBM’s overall HPC strategy, he noted. “The goal is to progressively engage clients in collaboration as a way to help us better understand different market segments and problem domains and to build better products… It’s about the right tool for the right problem” says Turek.

Even though this is essentially a collaboration rooted in research HPC, he stressed that the BlueGene piece isn’t the keystone of the story—it’s all about the data problems that are being addressed in novel, practical and workflow-conscious ways. “There’s a needed integration between big data and classic HPC technologies. These are inseparable concept for us going forward. Too often, players in the HPC space have cherry picked an algorithm or set of partial differential equations and they sort of thump their chests and say, ‘look how fast we made this go.’ The fact is, if in the general workflow of interest to the client, that piece of work went from a day to a second, you’ve just improved the overall performance of the much larger workflow by only a few percent.”

“Our industry has been hamstrung by self-defining itself as a vehicle to produce devices that try to optimally solve collections of partial differential equations or nonlinear/linear equations. But that can be such a small part of the overall workflow that constitutes an HPC workflow that we’ve ended up providing a degree of disservice to the industry at large.”

Turek says that when IBM introduced the concept of workflow a couple of years ago in their exascale conversations with the DoE and others, they made it clear that when you begin to look at workflows, it’s necessary to explicitly factor in data management, data flows and data organization as much as it does to pay attention to algorithms per se. When asked how this changes the direction for IBM’s HPC systems of the future, especially without an x86 play for some of their core HPC customers in government and academia, he noted that the data-centric architecture goes far beyond the micro-architecture view of evaluating systems. “You have to take into account where data sits, how its moved, and where its processed. Sure, there’s a processor involved, but there are a lot of ways to attack these problems from an overall workflow perspective.”

The final word from IBM is a steady firm commitment to HPC, a commitment that absolutely extends to the government and academic spaces (those areas being the focus of the question). “We are materially engaged with them in terms of co-design activities for future investments and our own investment in our Power servers to make sure that we’re offering them the best solutions in the world going forward. We’ll continue with these investments in HPC. We will be focused in terms of getting that investment centered on our Power technology and we’ll preserve our business relationship with Lenovo to provide the Intel-based technologies that customers we encounter might require.”