Blade Servers on the Cutting Edge

Each Blade server is an inclusive computer system that incorporates the processor, memory, network connections, and associated electronics onto a single motherboard. Because of their compact size, multiple Blades can be housed in a single rack or enclosure, which means that secondary resources such as power, cooling fans, and cabling can be shared across the entire rack. And because smaller servers consume less power and generate less heat than their bloated counterparts, more than 300 Blades can be crammed into a single rack - compared to an industry-standard rack that can only accommodate 42 servers.

With the current cost of data center space hovering at about $300 per square foot, not including the huge energy costs required to maintain a IDC, it’s no surprise that the cost-effective Blades are becoming increasingly popular. Add to that easier installation, dedicated software that improves performance and reliability, and the ability to replace a Blade server without having to shut down the entire rack and this technology seems to be an obvious home run. On the downside however, the devices are usually designed as “single-purpose” servers, which can have an adverse effect on scalability. So if you suddenly need more firewall protection and less storage space, you can’t simply reconfigure your existing Blade servers to perform the necessary tasks - you have to start from scratch.

The First Start Ups out of the Gate RLX Technologies of The Woodlands, Texas was one of the first companies to adopt Blade servers as a server solution. The company, founded in November 2000 by former Compaq executives, began shipping product in May 2001. Boca-Raton based OmniCluster, an IBM spin-off that began operations in May 2000, has had much success selling its SlotServer 1000 product. Another early leader is Egenera, which began shipping its BladeFrame product in September 2001.

The savings experienced by these start-ups has caused the industry heavyweights - Sun, IBM, Hewlett-Packard, Dell and Compaq - to take notice, and all are expected to have similar products available in early 2002. Compaq’s QuickBlade model is already in customer evaluation testing and will start shipping en masse shortly.The Market Opportunity Market analysis group IDC expects that about 50,000 Blade server units will find their way in the data centers of ISPs and web hosters by 2001. By 2005, Blades will account for over $4 billion dollars of the $102 billion server market, with over two million units expected to be shipped.
But it will not be clear sailing for all the players in the market, and casualties have already begun to surface. Hit by the downturn in the economy, Blade manufacturers are seeing a decline in server sales - which forced Ottawa-based Rebel.com Inc., and FiberCycle Networks Inc. of Los Gatos, California to close up shop.

But the industry as a whole should not be discouraged, because the savings offered by Blade servers are substantial, and benefits such as easier monitoring, management, and maintenance will ensure that this product will have a huge impact in the industry.

Egenera is one of the early leaders in the Blade server market. Susan Davis, the company’s Vice President of Marketing spoke to TopHosts.Com about this new technology and how the company’s BladeFrame product is changing the shape of the Web hosting and IDC markets.

Just like when you rent a house or commercial premises for your business, you have so many rooms, bathrooms, and floor space to use. In a web-hosting environment, your area is defined as disk space and network transfer.

Disk space is measured in Megabytes (MB) or Gigabytes (GB). Megabyte roughly means 1,024,000 characters and Gigabyte roughly means 1,024 Million characters. Imagine a character as one key on your key board. These amounts determine how many files, documents, or data you can have on your web site.

Network Transfer is also measured in Megabytes or gigabytes which determines how much data (how many of your files, documents or data) can be downloaded (transferred to) people accessing your web site. The more people, or the more data each person accesses on your web site the more data is transferred on the network.

The more disk space and network transfer you use the greater percentage of the web-hosting environment you are using – therefore the higher the rent.

Just as no office building and home is the same, neither is every web-hosting environment. Some offices have stairs, others have lifts, some houses have ensuites, swimming pools, and gardens, – and others do not. Web-hosting environments are much the same, some offer bare structures to do just the basics and others offer an array of features and facilities to help you do just about everything you could ever need or want. Some of the features and facilities likely to be offered are ranges of software to use, components, databases, and server side script processing.

The similarity of renting an office or home to renting space in a web-hosting environment is even more similar. With some buildings a gardener and/or a guard is available to look after the gardening or provide security. In a web-hosting environment, you have support people to help you do what you need to do on your web site to make it grow and there are server administrators to protect and secure your web-hosting space.

When you rent a building there are key parts needed to work or live in the space, like rooms, offices, kitchens, toilets, and bathrooms. In your web-hosting environment, you will find equally important components that are required to make the space workable.

TH: Egenera founded in March 2000 and started shipping BladeFrame in September 2001. What activity happened in the interim?

SD: It was all development. The development of the product really started in June of 2000. It was basically just about a year to develop the product and then we went into beta testing in July of 2001. So it was essentially one year from idea to product which was very very quick in terms of entire new technology.

TH: How is BladeFrame different from other products on the market?

SD: I guess the first answer depends on what you’re comparing it to. From the standpoint of comparing it to other bladed server technologies (that have been written about quite a bit), the real difference is that our product is aimed at a different segment of the market. So while many of the bladed servers that are out there - and some that are coming out from the major incumbent vendors - are really aimed at Web serving and front end applications that are typically uni-processor only, or maybe focused on low power. Our intent was to focus much more on mission critical applications that require much higher capacities of processing and memory, etc. So for example the blades - while they have a 1U form factor, will support two or four high end high end Intel server chips and up to twelve of the four-way blades - will support up to 12 gigabytes of memory. So they’re really intended for serious applications not just sort of a front end Web server type of environment.

SD: PAN Manager is what we call the secret sauce that works in conjunction with the blade frame. The product itself is called BladeFrame. PAN Manager is the management software which is really the key to what lets you provision cluster and reallocate resources within the system. So PAN Manager is both a GUI (graphical user interface), and a command line interface which can do all of the setup and provisioning of the system.
TH: How does the PAN Manager software system fit in?

So essentially what you have in this BladeFrame is really the equivalent of 24 traditional servers, which the PAN Manager software lets you cluster in any combination that you want, and allocate or de-allocate those resources from a running application on the fly. When I say 24 traditional servers, they are different because essentially what these processing blades contain is simply processing and memory. To make it like a conventional server you associate that with virtual network resources, as well as disk resources that live out on a storage area network. Once you combine these things together they have the look and feel of a traditional server.

TH: The system as I understand it goes for about $250,000. What market are you targeting? What challenges do you face?

SD: We are really going after high-end enterprises. Our typical customer in either the financial services [or] the service provider market is trying to either accommodate lots and lots of customers (and again for the service provider market it might be a 3 tier application) or requires lots of departments or more high performance computing types of applications. So the price is not just for the processing, but also includes other components that would typically live externally within a data center. All of the networking components: fiber channel, connectivity, network connectivity are really in the box. So when you look at the price and look at the total cost of ownership, you’re comparing it not to what a standard stand-alone server would be and multiplying that by 24, but taking again a bigger chunk of what lives in a data center and putting that into one system.

TH: So what would be the value proposition you present when you go out to meet a potential client given the downturn in the economy and cutbacks in IT spending?

SD: It’s really promoted on a number of factors. One is really server consolidation, and the ability to take lots of small servers that are sitting all over data centers around the world and consolidate them into one place. The problem with having all of those servers is the fact that your typical utilization levels on them are very low -and some of the statistics that we’ve gotten from our customers show that as little as 15 percent of their servers are actually being utilized, or 15 percent of their capacity is being utilized at any one time. So the ability to consolidate that and to allocate and reallocate that processing capacity on the fly as things change during the course of a day, week, month, or season means that your purchasing can really go way down. So for example if your typical utilization is 15 percent and using our system you could double that to thirty percent that means that you’ve now saved yourself the cost of half of the servers that you otherwise would have needed to purchase. If we can increase it by a factor of two or a factor of three then the number of servers you need to buy is reduced significantly. So being able to use what you pay for is one of the real major value propositions that we have, which is resonating well with our customers.

TH: Is there a cost savings factor that you would throw out there about how much money can be saved by employing the system?

SD: Well, one of our beta sites Credit Suisse First Boston was quoted in Network World, saying that he expected to save between half a million and a million dollars per rack of servers using our system.

TH: That’s huge.

SD: That is huge. Absolutely. And it really gets into a number of factors–it gets into greatly increased utilization, it gets into tremendous savings because we’re consolidating all of the IO (input/output) in the system. So it’s tremendous savings in terms of fiber channel ports and gigabit ethernet switch ports that you would otherwise have to have. So for example if you had 24 individual servers and you wanted redundant connections to a storage area network, then that’s 48 SAN (storage area network) ports and each of those ports may be $1,500 or $2,000 a port. Well by consolidating those down to through our system into 2 or 4 of them, then yes, you are saving the equivalent of 20 of them - let’s say times $2,000. And you can do the same thing on the gigabit ethernet side though the port cost is slightly less. They’re close to about $800 per port. So again taking things that would typically be out in the data center, consolidating it in the box means tremendous savings in terms of the capital outlay cost of those kinds of things.

TH: Now there have been a few casualties even though this is a very young market. For example Rebel.com and Fiber Cycle have ceased operations. They’ve obviously run into funding difficulties, poor initial sales, that sort of thing. How is Egenera dealing with that?

SD: Funding is not an issue. I would say the difference for us - I sort to hate to speak ill of the dead - is that our view on those types of systems and those types of companies were that they were really focused on solving one specific problem which was probably more of a “point in time” kind of problem. There are two things those systems did well - they consolidated lots of uniprocessors into a small footprint and they generated low power and clearly when power consumption in California, etcetera, was on the front page, it sounded like a really good value proposition. And of course they were focused on a market which was largely imploded, which I’m sure you’re well aware of in terms of the xSPs that were popping up - you know every day there were probably another twenty of them.

And really what makes Egenera different again is our focus is really on an entirely different market. The system is designed to solve a whole host of issues in large data centers and we’re focusing on the largely high-end enterprises - financial services is really our first because of course that’s where the company was founded and is where the experience of our founders lies. But the financial market also tends to be early adopters of technology, and is willing to spend money on applications that will drive revenue for them - which is where our product fits best. So we really didn’t get caught up chasing the xSPs (although we do have some customers in the service providers market), but we certainly expect that they will be among the survivors in this. And clearly a lot of the ILECs (incumbent local exchange carriers), will obviously survive, which is another good market for us as well.

TH: When do you think you’ll see actual traction? What are the early sales like or what are the adoption statistics like early in the game?

SD: Well, we believe we’re seeing traction already. We’ve had some very successful beta sites. A number of them are among very large Wall Street investment banks and we’re focusing a lot of our attention again at the financial services market and our sales are really ramping at the rate at which we were expecting.

TH: There seems to be an interesting battle with the Transmeta versus the Intel chip in this market. How are you involved in that debate or battle?

SD: We’re not. The Transmeta systems that were out there from some of the companies you mentioned were really focused on uniprocessor low power - which is not a road that we ever were going down. So we had very early on committed that it was Intel and the high-end processor road map that is our focus (the 2 and 4 way Xeon processors), and not the low power chips.

TH: Do you have any thoughts on how that’s going to unfold? Would Transmeta be in trouble?

SD: I can’t really comment on that. Obviously Intel has been an enormously successful company, and our view was that it was important to try and bring a lot of these high end features to a non-proprietary environment - to what’s become an industry standard which is Intel. And as part of that, we wanted to go with Linux because we felt that it was really hitting the tipping point in some of these large enterprise accounts, where people were looking at using Linux as way of getting off of the proprietary Unix operating systems which were really locking them in. So the nice thing about going to Linux/Intel is that there’s thousands of ISV applications that are now available on it which is clearly the key to the success of any operating system. That, and the price advantage of going with Linux led us to choose our current path.

TH: It also seems like the big five (IBM, Compaq, Dell, Hewlett-Packard, Sun Microsystems), use Intel products pretty much across the board, which is a huge endorsement for their chips.

SD: Right.

TH: Speaking of what the big five are doing, all of them are planning similar products - I think Compaq is in customer testing with the Quickblade. What impact does this have competition wise on companies like Egenera?

SD: From what we know so far - and we try to get as much information as we can obviously but there’s not a ton of it out there - the bladed server products coming out the incumbent server vendors seem to be following the path of some of the start ups. They (are) really looking at it as a low end solution, aimed at front end applications platforms for Web serving and that sort of thing. So they’re not bringing to it 2 and 4 way SMP (Symmetric Multiprocessing). In many cases they want to protect their proprietary Unix space, because of course the margins are so much better there. So they are probably not going to offer a lot of the management features that we offer with PAN Manager. So they really continue to be focused much more on front end applications that on targeting these kind of servers into their mission critical applications - as this is largely where their Unix systems have been deployed.

TH: I noticed that IBM is reselling RLX’s blade server product and they’ve also invested in OmniCluster. These are your biggest competitors, right?

SD: Not really, because RLX is focused on the low capacity, low power market - so is OmniCluster Really there are no other exceptions that we know of right now. Our competition is really the proprietary Unix system. It’s the RS6000s and the K Class from HP and certainly Sun Solaris that we are really competing against. And I believe the RLXs and Omniclusters and the rest of them are competing amongst themselves for what really is sort of a disappearing market.

TH: Do you think that the big five are looking at Egenera as a potential acquisition?

SD: I can’t comment on that. I really have no idea. It’s a good question for them though.

TH: Do you expect that there will be some consolidation in this market?

SD: The last I heard HP and Compaq were planning on consolidating. In terms of the startup though?

TH: Yes.

SD: I don’t know. Certainly this is a tough market for anyone, and anything is possible. Clearly our founders came into this business to build a business, and they’ve been very clear on that from day one. They brought this company into being through tough times. The fact that we successfully raised the money that we have especially in the last round a very, very tough market is a real testament to our plan, our strategy, and the amazingly quick product development cycle that we’ve had - and I think some other companies just got caught short. I think this is a great time for a “trial by fire” if you will, but we’re here to build a company.

TH: I actually don’t know about the history of funding. Who are some of the major backers?

SD: Our first $20 million was raised in October of 2000 and our primary venture capital investors were at that point were Spectrum and Kodiak along with Credit Suisse First Boston and Goldman Sachs. And then we raised another $30 million last July in again a very tough environment for VC funding - with those same VCs - and also added in an additional investment from a company called Austin Ventures out of Texas, and they led the round - it was really their first non-Texas investment. So again I think that was a nice confirmation of what we’re doing.

TH: What challenges do you think are ahead in the new year for Egenera?

SD: I think it’s really just executing our plan- which so far has gone like clockwork, in terms of the fundraising that we’ve done, in terms of generating investment, and in terms of product development. I mean it’s quite amazing to me that a year and a half ago some people sat down in someone’s kitchen with a blank piece of paper and said: “here are the problems we need to solve and here’s the system we need to build to do it.”
And a year later it was done, and in beta sites it’s worked very very well. Our beta customers have really been excited by it, and we plan to continue to move ahead and execute our plans. But it’s been amazingly smooth. I’ve worked for a lot of big companies with a lot more resources to bring to bear, and I’ve never seen anything go so according to plan.

TH: As a service provider or even an IDC provider, what kinds of benefits are there in incorporating your system?

SD: Well I think first of all one of the advantages that our product really has is that one of the issues that I know a lot of the service providers have today is that everybody sort of wants their own discrete system. They’re concerned about security, they’re concerned about making sure that SLAs are met and that their processing requirements are there when they need them - which is actually not very different from a large end enterprise who is serving the sales department, the marketing department, and engineering - and everyone feels like: “I need to have my own.” But one of the real nice features of our system is the fact that everyone can have their own, and yet from the point of view of the people managing it, it’s one system. So for example, with a fully loaded blade frame, you have this equivalent of 24 2-way or- 4-way servers which you can cluster together, or segment into what we call “logical partitions” or LPAN - each customer of the IDC has their own view of the resources of the machine. And you can even delegate the administration of those resources out to the IDC’s customer and they’re totally secure between them.

So you’re really solving the problem of being able to increase utilization overall, which thereby reduces your cost by giving everyone their own secure processing, resources, and management capabilities. So it’s a real nice fit into that environment. Clearly in terms of the industry challenges of finding and fighting over the existing customers - that’s more of an economic challenge. But I think they will find that our system will really help them reduce their costs. Our product allows companies to greatly improving utilization, and allows them to be able to reallocate processing from one customer to another as peaks are met, without having to buy to hit peak provisioning time. This combined with the physical simplicity of the machine, should be very attractive to potential clients. One of the statistics we’ve heard is that 25% of a system administrator’s time is spent running cables - which is just an enormous cost for what are typically pretty expensive resources. The fact that we’ve basically eliminated almost all of those cables means that you put the system in once and you never have to do it again. The blades that we offer all plug in and out of the system with no cabling whatsoever so they just slide in, they’re hotswappable, etc. So the management costs are reduced, the overall capital costs are reduced, and the utilization rates go way up, so it’s really just a great fit into that market.

TH: Is BladeFrame designed to compliment the bigger servers? I know you’re saying the finance companies are one of your major targets. I can’t see the need for those big servers going away especially for the banks or even the big E-business players.

SD: I guess I would answer that by saying that if you look, our systems are essentially through our blades 2 or 4 way server. Clearly there are applications that require many way SMP systems, 8 way, 12 way, 32 way and whatever, and there are systems out there that do that very well.
And that make a lot of sense but some of the other numbers that we’ve seen is that maybe 80% of the systems in these large data centers are 2 and 4 way servers. So they certainly don’t replace everything that someone’s going to have, but they certainly hit a big chunk of the types of systems that they’ve implemented today. And the other point that I want to make is that sometimes some of these real large many way SMP Unix systems are bought not because someone needs 16 way SMP for a particular application but so people can slice and dice them through the domaining capabilities that some of the incumbent vendors are offering. What they are really attempting to do is simplify the management by buying a very large system and cutting it into small pieces. The only problem with that is when you are doing that with an E10K (Sun) or something like that, it’s just a really expensive way to get a 4 way server. So if your application requires 16 way SMPs then those kinds of systems are perfect for that but if you’re really looking to buy something that will let you much more easily manage a collection of 2 and 4 way servers, our system is much more cost effective to do that.

TH: Are we just seeing the tip of the iceberg here? IDC and Yankee Group are putting the markets at about $4 billion by 2005 which is only 4% of the actual server market? What do you think?

SD: Well I think obviously a lot of that reflects just a general economic slowdown and the changes in the acquisition rates of systems themselves so it’s sort of hard for startups to have a view of that. We think those are great numbers and we certainly expect to have a good piece of that market.

Saving floor space and lowering power consumption have been touted as the major benefits of employing blade server technology, as customers can reap a 17 percent cost savings over a three year period. However, “the benefits of using blade servers go beyond maximizing real estate and saving on electrical bills,” says IBM’s Dr. Tom Bradicich, CTO of IBM’s eServer xSeries line.

Cable management is another big factor that’s worthy of consideration. A normal “non-bladed” rack is a maze of hundreds of cables, which “can be reduced to tens of cables with a blade architecture,” according to Bradicich. As a result, “there is less cost for cabling, less room for error by bumping and knocking cables about, and better access, with no need to go through wall of spaghetti to get to the server.” Eliminating all these cables may not seem like a big deal, but the impact on deployment is enormous - since blade servers plug into groups with few cables, they can be installed and configured in a fraction of the time of a traditional rack.Flexibility is another important selling point of Blade technology, due to the enhanced scalability they offer when it’s time to upgrade. “Once a company has invested in blade servers,” said Bradicich, “they will be able to take advantage of next generation technology without having to replace the box.”