Since 1986 - Covering the Fastest Computers in the World and the People Who Run Them

Language Flags
November 6, 2012

For DDN’s Alex Bouzari, Big Data is the Democratization of HPC

Michael Feldman

DataDirect Networks (DDN) is one of the few specialist storage vendors that has weathered the storm of industry consolidation. The company has managed to remain independent and focused on its HPC business, while others like BlueArc, Engenio, XtremIO, and Whamcloud got scooped up by larger companies with more diverse business goals. We recently spoke with DDN CEO Alex Bouzari to get his take on the industry churn and other trends that are reshaping the storage market.

DataDirect Networks CEO Alex Bouzari  

HPCwire: Alex, we’ve all noticed that the storage market has been changing very rapidly in the last few years — a lot of consolidation on the business side, and a lot of new technologies have been emerging, especially in relation to the big data application space. How has all this affected DDN?

Alex Bouzari: You’re absolutely correct. Big data is now here, although everybody’s interpretation of what it is, is different. The way we look at is that big data is simply the democratization of HPC. It’s basically the ability to deal, at scale, and in a distributed way, with large quantities of content, which is what HPC is. It’s a very positive development for HPC players.

With respect to the consolidation in the storage industry, we have found ourselves in this interesting position of not only being the largest private storage company in the world, but we are also now being tracked by IDC, in the top six lists that they maintain. It’s given us more visibility into customer environments, simply because the CIOs and IT managers who are looking at their options for big data, now notice us because we’re on their radar.

The flip side to that is that a lot of the big players — EMC and NetApp — are now talking a lot about big data. That’s creating an opportunity for people who are considering revising their IT infrastructure to look for alternate solutions. It’s been a very positive turn of event — the expansion of the addressable market — with the big data market becoming broader. And the publicity the big guys are now giving to big data, which, by the way, I think is going to benefit all the people in the HPC space — in fabrics and interconnects, in processors, and in storage.

HPCwire: So what kind of new platforms and storage architectures do you think are going to come about as the result of the big data application space? Is storage going to fundamentally change in some way?

Bouzari: Yes, we think it is going to change. Short term, it’s going to change by having some additional components inserted into the mix, which will make the ability to ingest, store and process big data easier, faster and more cost effective. Longer term we think the architecture will evolve.

Solid state memory is now getting to price points that make it attractive to some big data and HPC applications. Our SFA 7700 offering, which we’re going to announce at the Supercomputing Conference, is a hybrid storage solution that combines solid state and spinning disk, and does so for big data applications. It provides significantly better performance than an enterprise-type hybrid solution.

It comes down to the speed of solid state, in IOPS, the bandwidth of solid state, a large number of spindles, and the price point of low-cost spinning disks. All of that applied with some intelligent algorithms that tie the solid state into the spinning disks, results in a platform suitable for big data and HPC-type workloads. So we’ve taken a different approach architecturally to solving the problem of hybrid storage solutions.

The other aspect is that with all the developments happening in the world of processors and the significant investment Intel is making in enabling HPC to move faster, in our view, there will be a long-term shift in software being the enabling factor for heavy workloads like big data and HPC. And that software will gives organizations the ability to capture content from any source, store it in the most cost-effective way possible, either inside the datacenter or in the cloud, process it extremely rapidly, and make it available to users in a distributed way.

We think the nature of software as it relates to storing, processing and distributing content is going to change in a fundamental way. Processing power is becoming free; storage is becoming free. But what is not free is the ability to bring the content in, process it, and take it back out. That still requires huge amounts of infrastructure and software. So we think the commoditization of the hardware, which will take place over the next decade, will have to be enabled by really intelligent software that ties the pieces together.

We started moving in that direction three or four years ago, when we began embedding file systems inside our storage architecture. We’re taking it one step further by tying in solid state with disk. And now we’re actively working on next-generation architectures — three, five, seven years out — which are really aimed at solving this end-to-end problem where content can be anywhere and the users need processed content delivered anywhere.
HPCwire: Are you considering getting into turnkey appliances wrapped with Hadoop-type stacks or other big data software packages? Is that type of integration something that interests your company?

Bouzari: The SFA 7700, which is our hybrid appliance that combines solid state and disk, combines a management layer on top to give our customers something that is plug and play, easy to manage, and optimized for their big data workloads. It shows users what is going on — hot spots, what is being over-utilized, under-utilized, and so on. It also has self-correcting mechanisms.

The world is moving in the direction of appliances. In our case, we started going down that path as we broadened our market penetration in HPC, below the top 100 supercomputers. In the latest list we have 60 percent of those top systems, since we’ve always focused on the needs of customers at the high end.

But what is noteworthy is that we’ve increased the number of customers in the top 500 by 30 percent. So we now have a total of 150 customers for those systems. Many of the ones below the 100 mark are customers looking not just for the benefits of a high performance storage solution for their workloads, but ones that can be deployed very easily, since they don’t always have the expertise, budget and staff to do the things that the big guys are doing.

We think appliance-sizing what the big guys have enjoyed for quite some time with DDN, and making it available in the 7700 solution, will greatly benefit our customers in the HPC space, which is really the core of our business.

HPCwire: As you build these types of appliances, it seems like you be able to offer something that is more generally applicable across a range of application domains. Do you intend to stay focused on HPC?

Bouzari: The way we think of big data, whether its internet content creation and distribution, social media, digital security, genomics, oil & gas, or financial services, these markets are really an extension of high performance computing. So are we expanding in markets adjacent to HPC? Absolutely.

More than 50 percent of our revenue today comes from outside “core” HPC. However in terms of the requirements, the technology roadmap and what our engineers are working on is very much HPC-centric. The attributes that are required to do a good job outside of core HPC really have to do with things like ease of use, ease of management, appliance-sizing the solution, compatibility with the mainstream enterprise solutions, and optimizing with some of the enterprise databases.

But at the core, in terms of the ability to ingest lots of content, store it very cost-effectively, process it extremely fast, and make it available to users — whether it’s a big government lab or a firm on Wall Street — it’s very, very similar.

The main differences is that within HPC we have the benefit of having rocket scientists, who know everything and want to tinker with solution to optimize it themselves, whereas in the broader enterprise customer base, they still want all these capabilities, but they want it to be easy to manage, plug compatible, and certified for their environment. So we do that. But the bulk of our R&D spend is in HPC.

In fact, a year ago we kick-started our exascale initiative to develop the ground-breaking technologies required for that level of computing and storage to become a reality — a cost-effective reality. We’ve put in place a $100 million R&D effort to support that work.

So we’re very much HPC at heart. It just so happens that HPC is a much bigger market and it’s called something else: big data.