The server market has traditionally focused on bigger and faster machines, the Big Iron that runs large corporate networks and databases. But a number of companies, including Sun Microsystems Inc., are changing the way applications are deployed by networking many small, thin computers known as blade servers and by pooling their processing power.
Led by upstarts such as RLX Technologies Inc., some of the large system vendors have jumped on this innovation within the server market. Sun's Subodh Bapat, chief technology officer for Sun's volume systems products division, recently sat down with the IDG News Service to discuss his company's recent blade server announcements, and its vision for the next generation of the technology.
Q: What is Sun's approach to the blade server market?
Bapat: If you track the history of computing systems, you'll notice that once a decade or so, there is a major shift in the architecture paradigm in the way computer systems are built. We think there is a major new one on the horizon, and that has to do with throughput computing and horizontal scaling.
If you just look at where we've come in the past 10 years, we have come to an era where everybody takes a Web browser for granted, everybody does Web-based transactions, e-commerce is a reality, and applications servers that connect e-commerce front end with credit card databases and transaction databases and inventory database are now an accepted fact of life.
Over time, as more and more enterprises start deploying Web services to gain more efficiency in their business processes and more consumers start doing Web-based transactions, we think the name of the game is going to shift away from vertical scaling to what we call horizontal scaling.
Q: Can you give us an update regarding the blade products that you just recently introduced with the N1 announcements?
Bapat: We announced a variety of different blades. We're one of the first vendors in the industry to offer a mix-and-match approach with multiple different kinds of blades that all go into the same chassis. We have our SPARC (Scalable Processor Architecture) Solaris blades, we have our x86 blades that can run Linux or that can run Solaris x86. We also believe in integrating some of the networking value very close to this dense configuration of servers that you get in a blade chassis.
We announced a load balancing blade, which again is a standard blade. It comes in the same form factor as any of our server blades and fits into the same slot in the chassis as any of our blades. The same thing goes with the SSL (Secure Sockets Layer) specialty blade that we introduced. As you know, security is an important consideration for anybody deploying large systems anywhere. We've integrated the SSL accelerator into the blade chassis, thereby making security an integral part of the blade platform.
Q: Are you planning to release a new line of Intel Corp.-based blade servers in the coming quarter?
Bapat: Yes, we are looking at the next generation of our LX50 line.
Q: Now, do you plan to stick with the Pentium III processor for that? Or do you plan to upgrade to some of the newer processors that Intel has brought out, namely the Pentium M, which has some interesting power characteristics?
Bapat: We have a group here that is very active in tracking Intel's road map. We're on top of Intel's road map, and we're on top of AMD's (Advanced Micro Devices Inc.'s) road map. We have excellent relationships with both Intel and AMD. Intel's chip group considers Sun's systems group an important customer because we're one of the few three or four major systems vendors in the industry that is now committed to the Intel line.
I can't comment on the specifics of which specific processor we will use, we will make that announcement at the launch. We have locked down on our decision as to which one we will use, it will be very competitive when it does come out, it will be very aggressively priced, and it will be very performance competitive and price performance competitive.
Intel is vying actively as is AMD for our x86 business. The fact that Solaris scales better on large scale x86 systems than Linux does or Windows does, is something that is attractive to these x86 chip vendors as they try and go up the food chain and truly make x86 a platform to penetrate deeper in the enterprise. Linux and Windows, while they're trying to get there, they're not there with Solaris-caliber reliability, availability, and scalability just yet.
Q: What is the most important buying consideration for an IT manager considering a blade system?
Bapat: Budgets are flat, at best. So everybody in an IT department is looking to squeeze as much utilization out of their existing deployments as they can. One of the things they're looking to do is buy a blade server that runs the application I need at the time I most need that application, but that can also be dynamically on the fly reconfigured to run something else when I no longer have the workload that I used to have on my first application.
(They want) this ability to take a densely packaged set of server blades and create virtual domains or virtual boundaries within the server chassis, and segment it up using various virtualization technologies.
Q: Is that how you are positioning your first generation products, or is that what you have in mind for the second generation?
Bapat: We have this ability in our first-generation blade product, because our first-generation blade product ships with a version of our N1 software as an option. It will allow the retargeting and repurposing of individual blades for different applications. We see this as key to our horizontal scaling and throughput computing message going forward, because this change in paradigm would not be successful without the right virtualization, provisioning and policy management mechanism in the systems management area. Our vision of N1 is extending this not just to the blade level, but also the rack level, and the entire data center level.
When are your second-generation blade products due?
Bapat: The second-generation blade products are not due until late next year, late 2004. We're doing a lot of new innovation in those blades. For example, we are architecting them to run bigger applications with more memory and more compute capacity than our first-generation blade products. Our first-generation blade products are really targeted for deployment or as replacements for 1U rack servers.
The second generation is intended for a little bit deeper in the data center. It will have the ability to talk natively to SANs (storage area networks), to talk to Fibre Channel, as well as of course to LANs and Ethernet networks. The second-generation blades are where we're looking at changing the entire backplane fabric with Infiniband. We think that by making this system interconnect fabric an integral part of our next-generation blade architecture, we will bring tremendous cost reductions and tremendous value to the middle of the data center.
The blade server we released this year uses Gigabit Ethernet as a backplane fabric. Infiniband is a high-bandwidth backplane fabric that is explicitly designed for low latencies. We're talking tens of microseconds of latency for server to server communications.
It's also designed to handle block I/O to SANs and Fibre Channel networks. For a blade server that is targeted at running databases, or ERP (enterprise resource planning) or CRM (customer relationship management) applications, which is perhaps a little more heavy lifting, the Infiniband fabric is far more suitable for that. Our second-generation blade product will be focusing on Infiniband and we will be releasing a variety of different Infiniband-based blades that will fit into our second-generation blade chassis.
Because the architecture, at least Sun's architecture, has been elegantly designed in order to correctly partition the functionality that goes in the blade versus the functionality that goes in the chassis, it actually allows us to independently evolve the chassis, and independently evolve the blades. We can make a variety of different kinds of blades, all of which can connect to different kinds of networks, and that will plug into the chassis and draw their network services from the chassis backplane.
So you can actually have a blade road map evolving on a independent dimension than a chassis road map.
Q: Where do you see threats to Sun in the blade server market, from companies that are well-established like IBM (Corp.) or from newer companies that are trying to do some new and different things?
Bapat: All of these companies are trying to innovate in the blade area, and as always we take all competition seriously. We're obviously tracking closely what IBM is doing, as we always track closely what IBM is doing. And we're tracking Egenera (Inc.) and RLX and so on.
One of the things you'll find is that some of the current blade server vendors today don't nicely integrate the networking aspect into the blade shelf. Some of the companies in the blade market do not integrate networking switches into their blade chassis. So what happens is if you want to send data from one blade to its neighbor blade within the same chassis, you actually have to come out on a cable from each blade, out to an external switch, and come back up with a second cable into the neighbor blade. If you go to the back of the blade rack, what you see is cables coming out of the back of every blade, which really does not improve the total cost of ownership. It really makes it harder to manage and makes it all very complicated.
What we've done with our Sun Fire B1600 products is integrated and assimilated the Gigabit Ethernet switch as an integral element of the blade chassis, so you don't need to externally cable out to an external chassis. If you want to do blade to blade communications over Ethernet, it all stays within the chassis, there are no external cables involved.
As I said, we're one of the first blade vendors to actually come out with an SSL blade that fits into our blade slot, and a load balancing blade that fits into our blade slot. Everybody else is making announcements, but we've engineered it and we will be shipping very shortly.
Q: What do you see as the future of this type of technology? What types of things will be incorporated into not really your second-generation products, but your third or fourth?
Bapat: That's an interesting question. We're actually seeing that there are a number of applications that do well with horizontal scaling. There are a number of applications that continue to do well with vertical scaling. So we continue to think that applications that do require large amounts of memory and cache coherency will continue to be deployed on the vertically scaled architectures.
However, horizontal scaling is also going to gain increasing prominence over the course of the coming years. There's a very wide variety of application workloads that scale linearly or nearly linearly with that kind of architecture. In addition, you find that some of the middleware and database vendors that used to run on big memory systems are also rearchitecting their systems to run on clustered horizontally scaled systems. I'm sure you've heard of the efforts around Oracle (Corp.'s) 9i RAC (real application clusters) and some of the efforts around the CRM vendors and ERP vendors.
All of that is an attempt to move databases and software that required large amounts of memory on a large system to low-cost clustered building block systems, where the value comes not from the fact that this is a two-way Xeon server or four-way Xeon, or not that this is a one-way SPARC Solaris server or two-way SPARC Solaris server. The value comes from the fact that it has been intelligently clustered together with a high-bandwidth low-latency backplane and a management solution that can look at the rack as an entire solution.
It's the total amount of aggregate throughput in the system, and the efficiency of the systems and the reliability of the system as a whole that's going to count.
We're taking the interconnect technologies such as Infiniband and Gigabit Ethernet and offering that as a backplane technology for our next-generation horizontally scaled systems. We're making aggressive investments in our N1 management and virtualization software.
Dell (Computer Corp.) can't do this. Dell depends on somebody else for their operating system. They depend on somebody else for their systems management. They do not take a systems approach. They try to squeeze efficiencies from their supply chain, but they don't necessarily invest in R&D up the stack. And so that's one of the things that is going to be a differentiator for Sun.