Sun Microsystems Inc. is about to enter the second phase of its push into the competitive industry-standard server space with the rollout on Monday of the first two of its “Galaxy” family of systems powered by Advanced Micro Devices Inc.s Opteron processor. The first step was establishing a presence in the market, which Sun did with its V20z and V40z systems. However, the Galaxy servers promise better performance and improved management and cooling capabilities. Overseeing the development of these systems is Andy Bechtolsheim, a Sun co-founder who returned to the company early last year. Bechtolsheim, chief architect and senior vice president of Suns Network Systems Group, spoke recently with eWEEK Senior Editor Jeffrey Burt about the new servers.
Can you talk about what your goals were for these systems when you came back on board with Sun?
Bechtolsheim: From a Sun perspective, what we wanted to achieve was a common system image that would cover all these systems that we were [building]. Were only announcing the first two members—additional systems will be coming in the future. But the point is, all of these systems are essentially compatible, meaning theyll run the same BIOS, theyll run the same RAID controller, the same fault-management architecture, and this makes it much easier for us to support the range of systems. So we should support 10 different versions of operating systems—five of them are Linux, we support VMware, Microsoft Windows, of course, and Solaris. All of these had to be [quality assured] and validated against the underlying hardware, and thats just a lot easier if all of these boxes are identical down to the BIOS level.
The real reason we brought all this design activity back in-house instead of trying to find some boxes in Asia was to end up in an architecturally consistent state where we can have one release of Linux, Windows and Solaris that we know works on these machines in any given time and we can add some value in terms of the extra power [and] cooling service processor abilities that we have that are not generally available in Asia.
It was a combination of both adding value and being architecturally consistent. And obviously our customers expect that these systems fully work and work for all these different environments. Its basically a prerequisite for being successful in the market.
Have you accomplished these goals?
Bechtolsheim: Frankly, it took us a little longer to get the first set of boxes out than we would have liked, but we have many more boxes, some in the pipeline. What we really did at Sun was create an engineering process in terms of the whole [quality assurance] kind of environment that will allow us to bring other systems to market fairly swiftly. This is just an investment that Sun hadnt had before and it took us a while to get that set up. But were very happy with the systems themselves. I wouldnt know how to make them any different or any better, given everything we know today. As we look forward, we expect to see even more scalable versions of these products, which Opteron does support, of course, coming out that customers are basically waiting for.
You spoke about the faster Opterons—thanks to the higher power consumption—but can you elaborate on any other key differentiators between these systems and those of your competitors in this market?
Bechtolsheim: The main differentiation is the higher-power chips, which, unless you design the box for the extra cooling and power, you cant just plug them in. so that is something we think is going to be fairly unique for a while because, again, this chip was done specifically on our request.
In general, people like Opteron because of its performance characteristics, and theyll do whatever they can to deliver high levels of performance. Our choice was to really do this up front on these initial systems. … It is a competitive market, no need to say, and our choice was to focus on performance. Other people try to cram more chips into their power-limited boxes, which also works when you dont get the highest clock rate.
In other interviews, other Sun executives have used the term “radical design” when talking about the Galaxy systems. How would you describe the design? Is it radical?
Bechtolsheim: I wouldnt call it radical. We changed the industrial design so they look really nice and they look radically different than the previous Sun boxes, but radical is an overstatement. Were really catching up here with the enterprise-class from a functionality standpoint, but were also delivering the performance, so I think the combination of the functionality and performance will make us pretty well positioned in the market.
We also see a great roadmap for the Opteron ahead here, going to even higher clock rates and more cores in the future.
Can you talk about the development effort—how many people, how many hours involved?
Bechtolsheim: The total sense of the group that is doing the Opteron systems at Sun is about 250 people now, and theyre doing a total of about a dozen systems that are in development, of which we are announcing three at this particular announcement. So theres many more systems to come, including blade servers, etc., etc. We are obviously leveraging the Opteron design and the QA testing methodology on a wide number of systems and we intend to extend the line both upwards and downwards.
Can you talk about the relationship between AMD and Sun? This was much more than simply a vendor-OEM relationship.
Bechtolsheim: Theres actually more innovation and more investment theyre making on some of the initial systems that were not announcing today. … Those are boxes that are not similar to anything on the market and, again, that gets people excited.
We are very pleased with our relationship with AMD. They have certainly delivered on their commitment, and the high-powered chip is one of the results of that collaboration. We are closely engaged with them on future systems development as well, but we have no complaints whatsoever with AMD as a supplier to Sun.
Next Page: Opteron systems expand Suns market.
Opteron systems expand Suns
Analysts have said that Sun runs the risk of the Galaxy systems competing directly with its SPARC systems, particularly now since theyre targeted at the enterprise. Can you address that?
Bechtolsheim: We see Opteron primarily as a market expansion for Sun because we now can address Linux and Windows markets that we were not able to address before. There may be a small amount of substitution going on, but fundamentally I believe we were already losing those kind of customers that wanted to make a transition to the industry-standard architecture anyhow. With our Opteron box, we can capture them as part of the Sun product line.
SPARC is into the highly multi-threaded or throughput kind of architecture that we have discussed in public for quite some time. Those systems [based on the upcoming Niagara platform and UltraSPARC IV+ chips] will be coming out in the near future, and they also will have outstanding throughput and box performance and power performance, and in fact will have better power performance and throughput … than even Opteron had. Its a radical improvement on the SPARC side.
Now, what they do not have is the highest single-core performance. The Opteron, quite frankly, wins hands-down on the single core throughput, and thats important for technical, scientific, engineering applications, where you tend to run one program for a long period of time until its complete, and then the number of licenses that you need has to do with what the throughput per thread is.
When Sun first announced the Opteron systems, and later with the purchase of your company, was the talk about some internal resistance to the investment in x86. Have you run into much resistance in your work in developing the Opteron systems?
Bechtolsheim: What really changed for Sun when I came back was that for the first time, there was a clear commitment to deliver and to engineer systems in this space. I think the previous philosophy was that you could pick up these kind of white boxes or gray boxes in Asia and just bring them to market. I think what the company very quickly realized was that you dont add any value doing that. To deliver value to the customer, you have to do something thats essentially better than you can buy on the street corner in Hong Kong.
There is a difference between the capabilities you can deliver by designing it ourselves and what we can get from an Asian supplier. So even though all of this stuff is built and manufactured in Asia, these systems were fully developed by Sun.
You basically need critical mass in doing that, so … we now for the first time in the history of Sun have put a critical-mass effort on this effort, and that makes a big difference, including to customers because, if I was a customer, the first question I would ask is, “How do I know this is really compatible? How do you know if there is an issue, you can fix it?” And if you dont own the source code or dont know whats in the BIOS or service processor, how do you really do that? Were making a commitment here, a long-term commitment to the market for industry-standard architecture, and thats really the combination of the AMD chip in terms of its performance level and our ability to add some value that got us excited to get into this market.
Quite frankly, theres plenty of people supplying these type systems into the market. Its more fun to be different.
One of the mantras in the industry over the past year or so is bringing down the power consumption of the processors to enable enterprises to more easily deal with thermal issues. Yet you are bringing on systems with chips that consume more power.
Bechtolsheim: Opteron already is twice as power efficient as Intels [chips], and you get more than twice the throughput for the same power consumption as Intel, so were really ahead. And AMD is increasing the performance per power, so as far as we can tell here, yes, power is important—AMD has it now, Intel has it a year from now—so theres a major difference here. … If customers are concerned about power, the only solution that makes sense is dual-core Opteron. Its twice the power-efficiency-per-throughput than anything from Intel.