Suns Opteron Bet

By eweek  |  Posted 2004-08-23

Suns Opteron Bet

Sun Microsystems Inc. officials are betting that the combination of Advanced Micro Devices Inc.s Opteron processor and Suns Solaris operating system will enable the company to gain traction in the highly competitive x86 server space. Sun, of Santa Clara, Calif., is in the early stages of its strategy—according to Gartner Inc., the company shipped 490 Sun Fire V20z two-way systems in the first quarter, and Sun said that shipments went up 115 percent in the next quarter—but officials are pleased with the results to date. John Fowler, executive vice president of Suns Network Systems Group, is charged with growing the Opteron business, and recently spoke with Senior Editor Jeffrey Burt about the challenge.

When Sun brought out the Opteron systems earlier this year, the company separated the x86 systems group from the SPARC team. How has that worked out so far?

What Jonathan [Schwartz, Sun president and chief operating officer] was after was figuring out a way to accelerate SPARC development. As you probably know, the SPARC microprocessor team and the systems teams were actually separate prior to the reorganization. What David Yen [executive vice president of the Scalable Systems Group] has done is combine and reorganize the teams so that the microprocessors are actually together with the systems upon which theyre based.

On the Opteron side, the idea was that as a separate business unit, I could actually pursue my own pricing, my own product roadmap, my own marketing programs … without having a complicated mixed message. … We really see the Opteron product line as being a high-growth opportunity, so lets create a group to go grow it.

When Sun first began rolling out the Opteron servers, some officials talked about internal resistance from those inside Sun who felt the company should stay with its SPARC/Solaris roots and keep away from x86. How has that played out over the past six months?

The main resistance is really around cost. Obviously you have to go—as we have been doing—and invest in Solaris and Solaris optimization for Opteron as well as having another family of systems. I think that was the main concern that people had.

The enthusiasm of course is going up as we go and win customers.

So the early resistance didnt have an impact on Suns Opteron development? Are you on target with your product roadmap?

Things are going well. As you know, weve filled out our product line with a four-way [server] and workstations, and were working hard at expanding the product line out to include eight-way servers, as well as blade servers and other products that we havent talked about yet. Were going like gangbusters to fill out the product line. One of the great things about shipping in volume is then you build a bigger and bigger ecosystem around Solaris and ISVs and others, so youre just going to see more and more announcements from us after that. The important thing was to get started, and then build out the product line. Now were getting there.

Click here to read eWEEK Labs review of the Sun Fire V40z.

[Hewlett-Packard Co.] is rolling out a family of Opteron-based systems. IBM is offering Opteron in a system as well. Sun seems to be taking a different tack, though, working much more closely with AMD in growing the capabilities of Opteron to suit your needs. Can you talk about that?

Therere a couple of things going on there. First of all, we signed an agreement with AMD, a pretty broad-based agreement that isnt just a supplier agreement. What a lot of people dont realize is that we had been working on optimizing Java and Solaris for Opteron long before we had actually decided to build hardware products and systems products for that. In my old role as software [chief technology officer], I actually worked on the relationship with AMD as it related to Java and Solaris.

From a business standpoint, we have this great opportunity because everyone in the x86 world is going to be transitioning to 64-bit, and so its an inflection point by which we can take advantage of it by aggressively adopting Opteron—having a full product line—and then taking advantage of all the customers that want to go and upgrade from 32-bit Xeon. Its quite a different opportunity than it was afforded say a couple of years ago, which was only 32-bit Xeon.

In comparison to IBM and HP, we dont have really a Xeon installed base to go and protect or a big revenue stream there, so that we can go and price aggressively and upgrade customers from Xeon without confusing or mixed messages.

Next page: Combining Solaris with Opteron.

Combining Solaris with


What you do have is the development of Solaris. Can you talk about the combination of the two technologies—Solaris and Opteron?

On the technical aspects, obviously Solaris has been mature from a 64-bit standpoint, being the highest volume 64-bit operating system in history for a long time. So what were doing is a very large number of optimizations in the kernel and other parts of the system to take advantage of specific features in Opteron so that we run extremely high performance as well as have the reliability features that people are used to with Solaris. Some of that is shipping now in the 32-bit Solaris, and then we have a lot of that coming in the 64-bit Solaris, so we expect it to be faster and more reliable than any other operating system across a broad range of benchmarks on Opteron.

On the business level, the interesting thing is we own our own operating system, unlike all of the other vendors who are in fact shipping x86 hardware—that is Dell [Inc.], HP and IBM do not own an operating system, so we are already producing bundles by which the combination of pricing and service is 40 percent below what you would have to pay if you combined Dell with [Linux from] Red Hat [Inc.]. Its a very simple customer value proposition. Customers understand that buying the hardware is only the beginning of a journey. You actually need an operating system and a software stack above that, and you have to pay [for] service for those components over a long period of time. So with Solaris and with Java ES, we can combine those in promotional offers and subscription offers with Opteron hardware to get a much lower cost of ownership than our competitors can.

Can you expand on your plans on the technical side of the equation?

For example, we have N1 Grid Containers, which is a feature of Solaris 10, which allows you to run multiple application workloads on a single box without requiring you to have multiple operating system copies. … The reason this is interesting in the Opteron space is because as we build, in particular, the eight-way Opteron, people will be able to consolidate a whole bunch of applications on Solaris very easily on an Opteron box and manage those separate applications without the overhead of paying for extra operating system copies.

Weve done a lot of work at the kernel level to take advantage of Opterons memory capabilities as well as traditional registers for performance, and youll see those in benchmarks where running Java and Web services benchmarks, in particular, we will simply be faster than any other operating system in standard benchmarks running on the same hardware.

We see Opteron as a great way for building compute building blocks, very cost-effective compute building blocks. The long-run strategy is that people want to manage services. Today, services are often scattered among a bunch of compute building blocks interconnected by networking. What youre going to see us concentrate on is, in addition to coming out with bigger servers as well as blade servers, youre going to see us concentrate a lot on networking and manageability. We will have products to help with networking connections, and well also have a great deal more software, both as part of the boxes themselves for management as well as being add-on management software so people can be very cost-effective and efficient in how they manage these systems.

During the course of this year, well have additional systems come out that have a great deal of support for manageability, such as very flexible remote administration, complete remote upgrade, reliability and serviceability features that are not common in this price point—people can deploy very reliable systems with hot-swap components.

Reliability is a big topic. We have a middleware stack called Java ES [Enterprise System]; we license it at a very attractive offer on a per-employee basis. We include clustering as a standard component in Java ES. You can actually cluster and do highly available services without paying additional charges. As a general rule, across the product line, youll see reliability and manageability, both in the software and hardware components, be a constant level of effort for us.

We have a lot of experience in enterprise computing, both design of the computer systems as well as the operating systems and the middleware stack above them. Our heritage is really helping with the enterprise technology. What were doing is basically bringing that experience to x86, in both the hardware systems design as well as the manageability software and the operating system above it.

Next page: Sun-designed components.


-designed components">

Its the combination of Solaris and Opteron that enable this?

There are two aspects to this. Obviously we dont have to pay margins to someone else like [Linux vendor] Red Hat or [Microsoft Corp. for] Windows, so that we can just choose to combine the pricing and offer the lower price. From a service standpoint, since we create both the hardware and the software, we can very efficiently provide service to customers. If they have a problem, they call us up. Its all stuff we know. We dont have to have a complicated system for resolving problems, and therefore we can do it at a lower cost. From a business standpoint for us, its pretty simple—we can offer a lower cost to customers and we can still make money.

Right now this first generation of Sun Opteron systems are built with best-of-breed components. Youve said that over time, Sun will start putting more of its own technology into the systems. Can you elaborate on that?

What were concentrating on are elements of our own design where that adds value to the systems. So youll see new designs coming from Sun that employ different mechanical, thermal and power [features] that are of our own design [and] bring our own enterprise experience to bear, as well as support for support processor and BIOS, which again is where we bring our own experience to bear.

In the microprocessor itself and some other aspects of the system, its completely leveraged from companies like AMD, and we use today in our systems … Nvidia, and we use LSI Logic as other examples of companies that were partnering with. What youll see in systems going forward in the future is more and more components that we include our own design efforts on. It isnt willy-nilly—its only where we see we can actually add some value.

In the last [fiscal] quarter, Sun reported that shipments of x86 systems were up 115 percent [from the previous quarter]. Still, that was fairly insignificant to Suns bottom line. When do you anticipate Opteron making an impact on Suns financial picture?

We never give forward-looking comments about finances and all that, so Im going to stay away from what I think is material or not. But my plan is actually pretty simple, and the last quarter, which was our fourth quarter, was a good start on that. Im going to build up the product line, as we did with the four-way, and then going forward during the course of the year with additional products, and then were going to be continually working on different business offers as well as channel programs and sales programs to reach customers so that I can be growing regularly on a sequential-quarter basis.

The market itself is very big—the x86 server market is around $20 billion—so I can actually have a significant impact on Sun by only taking a small percentage of that because we are growing from a smaller place. My intent is to just grow regularly on a quarter-by-quarter basis.

Do you see that growth coming from within Suns installed base or from stealing market share from competitors?

Since Sun is in practically every Fortune 100 company [and] in a broad range of industries, it would be difficult to go to a customer that doesnt already have Sun. But [what] weve seen so far is that weve been able to go back and actually participate in business that we couldnt have participated in before. If the customer has already chosen x86, whether its for low-end price performance or because they want to run Linux or for whatever reason, the short list used to be Dell, HP and IBM, in whatever order [the customers] want, and we can now compete for that business. So far that has been entirely additive.

Next page: High-performance computing.


-performance computing">

Opteron has made its mark in the HPC [high-performance computing] space, but some industry observers are wondering when it will make its mark in the enterprise. What are your thoughts on that?

What we saw last quarter was a pretty broad-based interest. In the fourth quarter we sold our V20z—back then we only had our two-way server—to 324 customers in 41 countries. The leading customers were in high-performance computing /education, as well as financial services and telecommunications, so those would be sort of the classic three. We had representation from basically every industry—from manufacturing, government, software development, etc. So I think its going to take off.

And you have to remember, what AMD did [is] they made it possible to run all the existing 32-bit software, whether thats software on Windows or software on Linux or software on Solaris, so if youre an enterprise today, chances are youre running some of that software and you can use Opterons today, and it can run really fast. Its really a pretty simple proposition for the enterprise.

Where in Suns Opteron strategy does the acquisition of Kealia [Inc.] fit in?

Kealia is the name of a company that Andy Bechtolsheim was leading here in California. What Andy was working on—quite independently as a startup—was a line of Opteron servers, so when we acquired Kealia, we made a commitment to bringing those products to market. What were working on is finishing those products to bring them to market. Over the course of the next 12 months, we will bring various products to market. Theyre all Opteron products that Andy was working on as a startup.

So todays Opteron server products that were shipping right now, their development predated the arrival of Kealia.

Youve talked in the past about Opteron possibly growing beyond servers.

In general, the Opteron, because of its high memory bandwidth and big memory addressability, is going to be useful in many places, potentially also including storage. Were always looking at ways that we can use it, but beyond that, I dont really have much to say about that.

What advantages and disadvantages are there to jumping into the x86 market so late?

For us its relatively simple: Its a growing market—the revenue in unit volume growth, according to IDC as well Gartner, continues to go up in the x86 market. So obviously as a business you want to go and try to grow, which is important to us. And if you noticed, in our fourth quarter we finally managed to turn around: We had huge unit volume growth as well as revenue growth for Sun as a whole, so we want to continue to do that with all our products as well as x86.

The important thing is, being last to the market may not be as big a disadvantage as people think because we are up on the heels of the 64-bit transition and the emergence of Opteron. That gives us a great opportunity that we didnt have before to get on the ground floor and be a leader in this new area, which is going to be 64-bit x86.

What impact will Intels jump into 64-bit x86 computing [with its EM64T 64-bit extension technology] have on Suns business?

Its great because a lot of the Solaris work that weve already done can be leveraged on Nocona. Solaris and the compiler work, as well as the middleware stack, so I think its generally good news because it validates what AMD was doing. I would actually hate to be just alone with AMD 64 architecture because in general thats not really good for competition.

Does Sun have plans to broaden its Intel offerings to include Nocona or future 64-bit Intel chips?

Right now we havent announced any plans to do anything with Nocona. In general, the direction, with EM64T, is something that we applaud and were definitely watching it closely and if we think it will support a great value for customers, we would look at adding it.

At least right now, the Opteron, which you can build one- through eight-way with a single chip set and a single software effort is a much better value.

But youre not closing the door on any options.

No, were pretty pragmatic about microprocessors. Opteron is superior and we have a great relationship with AMD, so were very focused on Opteron. It will take a while for Intel to catch up; they have a long way to go to catch up on performance, and its going to be good. Competition is good.

Check out eWEEK.coms Infrastructure Center at for the latest news, views and analysis on servers, switches and networking protocols for the enterprise and small businesses.

Rocket Fuel