IBM Exec Lays Out Intel Server, Blade Strategy

With the launch of a four-way Intel server that leverages a system architecture reminiscent of a mainframe, IBM is making a concerted effort to differentiate itself from rivals such as Hewlett-Packard and Dell. Tim Dougherty, director for eServer BladeCenter Systems for IBM's Systems and Technology Group, explained the company's Intel server strategy and relentless commitment to blade servers in an interview with Editor In Chief Michael Vizard and Editor Heather Clancy.

CRN: In a nutshell, what is IBM's server strategy for competing against HP and Dell?

DOUGHERTY: Our big play is blades on one end and eServers at the high end. The squeeze in the middle is what has been the commoditized servers. If you compare a 1U server from Dell with one from IBM, you'll almost always find that we'll give you more DIMMs, more drives and innovation around vector cooling across the entire product line.

CRN: How does IBM's new four-way Intel server further that strategy?

DOUGHERTY: We think this is a trump card to play against HP. It uses mainframe technology to give us a significant advantage in the four-way space. We've spent $100 million developing the X3 architecture over a three-year period.

CRN: Beyond the traditional server space, how do you contrast IBM's approach to the blade server market from HP's?

DOUGHERTY: My view of HP is that they view blades as 1U servers thrown on their sides. If you look at their blade server equivalent, there are 112 fans in it. We have two. Those kinds of issues tell me that I'm not sure they quite get it yet. Our Cisco switch is the size of notebook. Their switch is another blade that you have to stick in. From a design point of view, they have a lot of work to do.

CRN: How can you get by with only two fans?

DOUGHERTY: Our blade centers will operate in excess of 90 degrees Fahrenheit. When these fans start up, they sound like aircraft engines. They're not really fans. They're huge blowers. One of them is enough to cool everything. We call it calibrated vector cooling.

CRN: What about Dell in this space?

DOUGHERTY: We have not seen a lot of Dell yet. If I were in their shoes, the first thing I would do is go to my installed base. If they do that, we wouldn't necessarily see them. Part of it is that I don't think they have their product right yet. They don't do much integration. They have a lot of plans, but they don't have basic things like N&#043N power. People said blades initially were just a form factor. But blades offer you the capability to do a lot more. We spent a year-plus talking to customers before we ever brought this to the marketplace. The thing that is attractive about blades is not the density of the servers. It's much more about how you manage the total infrastructure. There is a capability to reintegrate things that over the years have become disaggregated. That's one of the reasons we think blades will not be a reasonable success for Dell.

CRN: How does Sun Microsystems stack up as a competitor in blade servers?

DOUGHERTY: A significant portion of our blade sales have been at Sun's expense. Sun has a huge installed base that we've been picking off like crazy when they go to a Lintel base. Sun's blades don't make much sense, and now they are saying they will have a 2006 offering. They are using an old design today aimed at the edge of the network. Customers want blades to run business applications, not just edge-of-network kind of stuff. CRN: How do you see that change playing out in terms of how solution providers sell blade servers?

DOUGHERTY: This approach to integration leads you to much more of a solutions sale. We've announced that we're building a series of solutions. They are not hard bundles per se. They are soft bundles built around what we call a reference architecture. We are working on bundling with Citrix, VMware, SAP and some other unique things we can do like Retail in a Box. On the SMB side, you could have one that is Microsoft in a box and one that is Linux-based. This is predominantly a channel play because at the end of the day, they deliver the bundle that they can customize. The Avnets of the world are building solution centers to create those kinds of solutions on behalf of customers.

CRN: Some people would argue that IBM is trying to lock customers into a proprietary blade-server architecture in the Intel space.

DOUGHERTY: That's not new news. Once you buy the chassis, you're sort of committed. It's the only one that can run IBM stuff for now. But once you fill the chassis, there is no reason why you can buy your next one from HP or Dell. Industry pundits are asking for a chassis standard, but we have not heard customers ask for that. In fact, customers don't want to put an IBM blade in a Dell blade center or vice versa because they don't know who to call if they have a problem.

CRN: Given that, would you say the blade center market is open?

DOUGHERTY: Last October, we announced an open specification. We published the specification for how to build a blade, a switch and the connector that goes on the blade. You can build a blade that has any kind of function. This has been a collaboration effort with Intel from the beginning. They take our blade center with their ID on it and sell it to other OEMs. There are 40-something companies that offer a blade center.

CRN: Do you expect vendors that offer appliances for things like security to adopt a blade approach that would plug into a blade center?

DOUGHERTY: That is one of the reasons we opened the specification. A lot of the security space is based on appliances that sit in the network. It's clear that customers want to plug that functionality into the blade center. There are really three layers of things you can plug into a blade center. There are base infrastructure line networking and security. There are applications like PeopleSoft, and then there are actual vertical solutions like Retail in a Box. You can dream about almost anything. We take Ethernet, Fibre Channel or Infiniband switches and integrate them into the blade center. Our competitors are starting to do the same thing. But usually, they only pick one kind of switch, and that's not what customers want. They want to make the choice so we support all switches. Customers have already developed a whole set of standards on how they govern switches.

CRN: Do people buy the switches from IBM?

DOUGHERTY: Even though our networking blade from Cisco carries their label, you buy it from IBM. We provide the first- and second-level support. For all intents and purposes, it's an IBM switch. We do the same for Nortel.

CRN: What are you doing to drive adoption of blade servers in the channel?

DOUGHERTY: We have specific programs in the last six months to get more education in the channel. It's an iterative education process to show them they can make a lot more money. They may need to gain some additional skills in networking or storage, but the payoff is huge.

CRN: Are you trying to recruit additional resellers in this space?

DOUGHERTY: For the most part, we're not trying to recruit new people. I'm trying to educate the people we have. There are some new people that have come in, but for the most part it's an education process for my current channel. A good 70 percent of our sales are going through the channel, so we're doing reasonably well. CRN: But most people today perceive that blade centers are a high-end enterprise play, rather than something that can be sold at the SMB level. So it would seem that there is a lot of business to be driven in that space.

DOUGHERTY: Thirty percent of our volume is in SMB. Blade centers are positioned as an enterprise data center play by the way we came to market and the way our competitors positioned their products. The perception is that you have to have a whole bunch of servers for this to make sense. But a lot of customers are only putting two or three servers in a blade center. It's worthwhile because of the money they save on infrastructure management. Blades don't come with a lot of cables and all the other crap you have to worry about. You slide them in, and it can be managed remotely. You can ship it out, and all somebody has to do is put in the blade.

CRN: What challenges must solution providers meet to successfully sell blade centers?

DOUGHERTY: A blade center is an infrastructure sale. That means it has a longer sales cycle and more complexity. If they do it right, it's much more profitable. Our blade center chassis is $2,500, but we have blades that cost as much as $20,000. With a few of those, you're talking about a $100,000 sale.

CRN: Some people would suggest that server virtualization is a better approach than just throwing server blades at distinct applications.

DOUGHERTY: People talk about virtualization and server utilization. But the reality is that when most customers are paying less than $2,000 for a blade, they don't care. Over time. this will change. But the reality today is that this is not an issue for most people. The ability to plug stuff in and set the infrastructure once--including networking and storage--in a single box that eliminates 86 percent of all the cables is what they want.

CRN: What sparked all of the interest in blade servers?

DOUGHERTY: On the data center side, there are huge issues today. When the 2000 and 2001 [economic] crunch hit, individual [company] departments said they couldn't support distributed computing because they let people go. So they wanted to give all the servers back to the data center, but the data center didn't have any room to put this stuff in. It's not even a space issue; it's about how much power and air conditioning they can provide. You can make a lot of money on data center design these days. Customers have lost that skill. They use to have that skill 15 years ago. But it's all directly related to the Intel road map. As little as two years ago, the average processor was 30 to 40 watts. Toda,y it's 115 to 150 watts. That directly relates to how much air conditioning you need to provide.

CRN: At the end of the day, how do you sell all this in line with IBM's On Demand strategy?

DOUGHERTY: At some point in the future, you will be able to attach a blade center to a fabric, and that fabric will essentially deliver all of the I/O for you. Where we are headed is toward the concept of a common workload manager.