EMC’s Comeback in Server-Side Memory: Q&A with Pat Gelsinger

2/7/12Follow @wroush

In enterprise data centers, servers and storage go together like hot dogs and buns. One isn’t much good without the other. But if your specialty is baking buns, you probably don’t spend a lot of time thinking about how to improve the dogs. And that, in essence, is one of the limitations that Hopkinton, MA-based EMC has been trying to overcome lately.

The Hopkinton, MA-based company (NYSE: EMC) is one of the world’s leading vendors of storage arrays for enterprise data centers. Because it has always thought of itself as a storage company, it has never crossed the line into building components for servers. And that’s how, even though EMC was first to market with Flash-based memory technology for enterprises, a much smaller company, Utah-based Fusion-IO, was able to come out of nowhere in 2005 and take the lead in a burgeoning new market for solid-state Flash memory chips for servers. Today HP, IBM, and Dell all put Fusion-IO’s ioDrive cards in their servers; the company went public last year and is valued at around $2.1 billion.

But it’s a natural market for EMC, and it wants in, badly. At an event yesterday in San Francisco, EMC took the wraps off a competitor to ioDrive called VFCache. It’s basically a card full of Flash modules that fits into a “PCI Express” or PCIe slot in a computer server, where it provides an instant memory upgrade. That allows the CPU to get work done faster, without having to slow down to wait for data from the storage array.

EMC's Pat Gelsinger introducing VFCache, aka Project Lightning

How EMC came back from behind in the server-side Flash business is an interesting story of internal innovation. Pat Gelsinger, president and chief operating officer of EMC’s flagship Information Infrastructure Products division, told me yesterday that to help EMC catch up with Fusion-IO, he authorized an unusual skunk works project, with most of the engineering team isolated in EMC’s facility in Tel Aviv, Israel. “We hired the very best people and treated it like an internal startup,” Gelsinger said. “We set incredibly focused goals for the team. We told them not to go to corporate meetings,” but to concentrate solely on “Project Lightning,” the code name for the VFCache product line.

That was the only way to get the job done once EMC had determined that it had to build, not buy, its own alternative PCIe Flash product, Gelsinger says. “Organic innovation is very hard in a big, successful company,” he says. “There are a lot of antibodies saying, ‘No, you can’t do that, we can’t go that fast.’ There are a thousand reasons these things can slow down in a company. So having a very hands-on, top-down focus, combined with a very maniacal, senior, aggressive team, is really what’s required.”

Flash memory, a technology pioneered by Toshiba, is far more expensive than hard drive storage, but it also works much faster. Companies have begun to put extra Flash-based “tiers” of memory between servers and storage arrays in order to address the gap created by ongoing improvements in CPU speeds. Hard drives just can’t read or write data as fast as today’s multicore processors can suck it in and spit it out, which means CPUs are often sitting idle, waiting for data to arrive.

To speed things up, EMC has been adding Flash memory to its storage arrays since 2008—in fact, it has shipped more Flash memory to enterprise customers than all other vendors combined. But before Project Lightning, it hadn’t tried putting Flash into servers themselves, where the benefits are even greater, thanks to proximity to the CPU. Gelsinger says the VFCache devices bring about an “order of magnitude” improvement in server performance.

The team behind Lightning is already working on a sequel called, naturally, Thunder. EMC gave a sneak peek of the technology at the press event: it’s a network appliance that combines several VFCache cards to further improve communications speeds between servers and storage arrays.

My interview with Gelsinger, which is included in edited form below, also hit on questions about EMC’s big-picture vision of its role in enterprise computing, efforts to integrate recent EMC acquisitions like San Mateo, CA-based Greenplum and Seattle-based Isilon Systems, and the company’s relations with the venture and startup communities. I spoke with Gelsinger shortly after the company’s glitzy VFCache launch event, which attracted about 50 media, analysts, and bloggers and about 3,000 live webcast viewers on EMC’s website. (You can catch a full replay here.) The company played the Flash theme to the hilt, even appropriating the Queen theme song from the 1980 film Flash Gordon for a tongue-in-cheek “Cache Gordon” video that likely left Freddie Mercury rolling in his grave.

Xconomy: EMC was first to market with Flash-based enterprise storage products, but Fusion-IO grabbed the lead in this new market for Flash-based memory cache products in servers. How did that happen? Does today’s announcement represent EMC’s effort to catch up?

Pat Gelsinger: With Flash in the enterprise, EMC was the leader, is the leader, and we’re pretty committed to continuing to be the leader, period. In the segment of putting Flash in the server, Fusion-IO was the first. In that subcategory, they were first. I respect that leadership. We should have done that sooner. In that sense, we are playing catch-up, for that sub-segment.

But we’re the market leader overall. A big company like EMC is not going to lead in every particular area of innovation. For us it’s very much about the whole set of innovations. And as we went and talked to our customer, what we found in this subcategory of server Flash was that they were compelled by the performance characteristics of Fusion-IO but dissatisfied by its inability to fit into the overall storage hierarchy. It didn’t have protection. It wasn’t part of the same management environment. It was this island of Flash that wasn’t persistent with the rest of their storage investment. And in that, we saw the opportunity.

We have some of our cultural views. We look at the world very much through the lens of a storage array. We did have to sort of leap outside that and say, no, it’s about storage anywhere in the hierarchy, whether it’s on the server side or what. So that was breaking a bit of glass internally. We can be on the other side of the wire. As we did that, we said hey, we can extend our FAST [Fully Automated Storage Tiering] technology to the server. We crossed that bridge from enterprise storage arrays to a broad use of storage with the FAST technology, and we found we could really popularize or mainstream the use of Flash in the server with the extension of that technology base. That’s very much what the story is about.

There are lots of tech startups in this space. Flash is a very disruptive technology, with 1,000x to 4,000x kind of improvements possible. Lots of companies will try to innovate with that. But today EMC is unquestionably the leader in terms of delivering enterprise Flash. The 24 petabytes we have delivered, we believe that is greater than the sum of everyone else put together. Our overall share of the market in storage is 30 percent, and we are 60 or 70 percent of all enterprise Flash. So we are just way ahead of the market in that sense. We think with today’s announcement, we have covered the waterfront. We will have Thunder as a server network appliance for different use cases. What we are doing on the array side with our FAST technology stretching end to end across it, it really is a powerful combination.

X: What was the hardest thing about getting here, about creating Lightning? What technical or cultural problems did you have to overcome?

PG: We as a company have done most of our most innovative things by inorganic innovation—e.g. acquisitions. In this case, we looked at and said, there aren’t that many choices of companies to buy in this space. So what we really did was organic innovation. We took Mark Sorenson, the head of the unit, Danny Cobb, who you saw today, we built up an engineering team which is mostly in Israel, and we treated it like an internal startup. We set incredibly focused goals for the team. We focused them incredibly narrowly. We told them not to go to corporate meetings. “You’re not allowed in this room, get out of here, go back to work.” We hired the very best people, and we really treated it like an internal startup in that way. And I micromanaged it like crazy.

And there are two sides to that. You can say, “Micromanaging, isn’t that terrible.” The other side is that organic innovation is very hard in a big, successful company. There are a lot of antibodies saying “No, you can’t do that, we can’t go that fast, we need a longer this or that.” There are a thousand reasons these things can slow down in a company. So having a very hands-on, top-down focus, combined with a very maniacal senior aggressive team is really what’s required to do a successful organic innovation like this.

X: Are you leaving the same structure in place for Thunder?

PG: It’s the same team. Think of it as the second product from the internal startup. I expect that 95 percent of the code, the software that’s in Lightning, will end up in Thunder. There will be additional software required, but the vast majority of the technology required is Lightning. Essentially, it’s a group of Lightning cards coming into the server network Flash appliance. In that respect, it’s a bunch of Lightnings cooperating together inside of a network configuration.

X: What’s your larger, long-term vision for EMC? Where do you think you fit, how big can the company be?

PG: At the highest levels of IT, we look at four major trends going on: social, mobile, cloud, and big data. Those are the big four—you can talk about lots of others as well, but most of them are somehow mashups of two or more of the other trends. EMC is clearly sitting in a firm if not leadership position, in cloud and big data. Social and mobile are more consumer-oriented trends, and we don’t have consumer DNA. We don’t pretend to be a consumer company. We are an enterprise company. Our focus is on cloud, and big data, and trust. So those are the things we’re going to do. You look at our simple logo—where cloud meets big data on a foundation of trust. And that is what we want to do in the future—be the leading company for private-public- hybrid cloud environments. We want to be the biggest company with respect to leading the transition from structured, transactional database environments to structured and unstructured, predictive, big-data analytics environments, to be able to store those, to be able analyze those, protect those in the future, all on a foundation of trust and security. We are by far the leader in virtualization, in storage, and we’re trying to expand that portfolio, taking RSA, expanding its presence, increasing our software presence, and stretching our data center offering with things like V-Block and Lightning, and going increasingly higher in the stack, with things like Greenplum.

X: My next question was going to be about Greenplum and also Isilon. You bought Greenplum in San Mateo in July 2010 and you bought Isilon in Seattle in November 2010. How is the integration of those companies into EMC going?

PG: On Greenplum, it’s working well. Greenplum is not a storage array. In that sense we are having to learn some different skills. We would sell lots of storage into Oracle or SQL environments but that didn’t make us database experts. We are having to build and expand our capabilities in that space, with new partnership, et cetera. We grew greater than 300 percent in the Greenplum business in the last year, so it’s a pretty heavy growth area. And have clearly set a very aggressive cadence for the expansion of our strategy. We acquired the company and 90 days later we announced our first appliance. We delivered our first Hadoop distribution shortly thereafter. In November we introduced the unified analytics platform. I like to call it the integration of Facebook, Google, and Amazon for enterprise big-data analytics.

What do I mean by that? Google operates at full Internet scale and almost real time. With today’s data warehouses you put in a query on Friday and you wait until Monday morning to get an answer. We believe you put in a query and you want a response in 200 milliseconds. So it has to be interactive like Google. And operate not on subsets of data but the full data. It has to be like Facebook in the sense that as these data sets get to be so big, you have to enable people to collaborate on them. I might be able to give you a separate copy of a 6-terabyte database, but if I have 100 petabytes I can’t give you your separate copy to go play with. It has to become a collaborative environment. And we think the killer app becomes predictive big-data analytics. It’s this idea of Amazon saying “You might also like…” It’s being able to make real-time, current, collaborative predictions about what the next thing or the important thing would be. We think in terms of that, we’re doing very well. People are giving us full credit. I saw an article just this morning. Who are the two corporate leaders in terms of Hadoop? IBM and EMC.

X: Did you feel like you had to acquire a Greenplum in order to become experts in the hot database markets of the future?

PG: Yes. We didn’t have those genes ourselves. We looked across the industry, and this was a call I made personally. I saw the Greenplum guys as having the right cultural environment to join EMC, having the best technology—scale out, shared nothing—and we are in a very hot position in the industry, so we’re quite excited.

X: What about Isilon?

PG: Isilon is scale-out storage environments. It’s basically been very successful in media, oil and gas, genomics, medical, et cetera. Now we’re expanding into enterprise use cases, where the EMC sales force is very strong. We’re also doing more to integrate it. The examples of that would be, two weeks ago we announced Hadoop support of Isilon, so we’re integrating Greenplum Hadoop directly into Isilon arrays. Today what happens is, imagine you are using Hadoop to filter Twitter and Facebook and all this stuff and landing it into your Hadoop distributed file system environment and you want to combine it with customer record data. How do I cross the structured to unstructured worlds? The only way now is to copy it into some structured analytic environment. Now what we just did is we said, you don’t need to do that. The best way to land your Hadoop data is the best way to land your structured data. We’re going to run our Hadoop Greenplum analytics directly off the array. That’s an example of tying together our big data assets. I think

X: Speaking for the Boston entrepreneurial community—which Xconomy is part of—EMC is sometimes seen as a bit distant, quiet, and uninvolved. What do you think EMC’s role should be in supporting the community of startups and entrepreneurs in Boston?

PG: I don’t feel like EMC does enough in that respect, period. It’s not a Boston statement, or a Bay Area statement, or a Tel Aviv or Shanghai statement. So we have initiated EMC Ventures—-we had been doing venture investing for a decade, but we formed EMC Ventures a year ago so that we could have a more structured engagement with the venture community. We are also formalizing our university research program as well, so that we have structured engagements with the Stanfords, Berkeleys, MITs, Harvards, and UWs. You will see us do more in a systematic, structured way in those innovation communities.

At the opening of a research center at the Technion in Israel recently, I gave a speech about the “golden triangle” of innovation. A company that can successfully operate across the three corners of this triangle—the venture community, the corporate community, and the university community—will have sustained innovation.

X: But what about the startup community? Are you saying entrepreneurs will start to bump into EMC people more often?

PG: The simple answer is yes. And with that, we are doing more new college graduate hiring. EMC had been drifting toward hiring more experienced people versus a mix. We are also expanding our internship programs significantly. My view of interns is that you should have an aggressive program, pay them pennies, work them like dogs, and then hire the good ones. Your rate of success when you hire interns is much better, and you can be much more selective. So we are strengthening internships and partnerships and ventures, on a global basis but all of that applies in Massachusetts as well.

Wade Roush is a contributing editor at Xconomy. Follow @wroush

By posting a comment, you agree to our terms and conditions.