The Art of the Operating System

How to make innovation and stability coexist.

By Al Riske

01.Sept.06--If you want to know something about the Solaris operating system, ask Tim Marsland.

Marsland has had a hand in every Solaris release.

He joined the company when it was working to unify the BSD and System V versions of UNIX -- before SunOS became Solaris -- and has been writing code, fixing bugs, setting the direction of the architecture, and influencing the business strategy ever since.

"Over the past 15 years, Tim's technical leadership and innovative thinking have been evident in the extraordinary role he has played in developing and enhancing the Solaris operating system," says Sun CTO Greg Papadopoulos.

Most recently, with Solaris 10, Marsland designed the Solaris Express release model, contributed to the early design work on Zones, and played a critical role in the revival of the Solaris on x86 platform.

Now a Sun Fellow (the company's most prestigious engineering rank), Marsland is also widely recognized for playing an instrumental role in open-sourcing Solaris. An early advocate of the strategy, he worked to remove some of the related legal encumbrances before shifting his focus to Solaris on x86.

Ask Marsland about Solaris and he's most likely to talk about open source and x86. He's also likely to downplay his own contributions.

"Sun is an amazing place to work. What makes me look good is everybody else. It's hard to describe the depth and breadth of other individuals and how much I rely on their knowledge," he says.

What they all have in common is a desire to make things people will want to use, things that will (all modesty aside) make the world a better place.

In Marsland's case that blend of practicality and idealism is why he left his post at England's esteemed University of Cambridge.

"I had been very impressed with the things Sun was producing and I thought it would be an interesting place to go," he says. "I wanted to get into development and how stuff is really done, because I always saw myself more as a engineer rather than a pure researcher."

Marsland can name a great many bright spots in the history of Solaris, but the darkest day? No contest. It was the cancellation of Solaris on x86 in early 2002.

"That was a very bleak day for us," he says. "Most of the engineering world inside Solaris saw that as a possible path to volume and adoption, and we could see our lunch being eaten by the open-source world out there busily taking over the x86 space."

Still, Marsland and his fellow engineers couldn't let it die.

"The engineering community kept Solaris x86 going even after executive management decided it was the wrong thing to do. Very naughty, but we did it," he says.

"Our managers had the same contrarian mindset because they were allowing us to go on. It wasn't like they didn't know about it. So people kept it working. It wasn't that there was any investment going on, but we were deliberately not breaking it, which is cheaper."

That and a well-timed stunt were enough to turn the tide.

"The thing we did, which was almost a party trick but was very effective as it turned out, was ... a bunch of us engineers worked on making Solaris, release 9 at the time, come up on the first Sun-branded x86 box," Marsland begins.

As originally planned, the Sun Fire LX50 was to run Linux but not Solaris.

"The Solaris organization was going, 'Run Linux all you like, but why isn't Solaris available, too?'" he recalls.

So, the party trick.

"I captured the OS bring-up log in a piece of email and sent it off to Scott McNealy and Jonathan Schwartz, saying, 'Look, we got this thing going.'"

It seemed like a big deal but wasn't really, Marsland says, given the work that had been going on behind the scenes. He lauds Schwartz, who had taken over as head of software at the time, for running with the results.

"That was the rebirth of Solaris x86, because Jonathan was taking it very, very seriously. He understood the volume and adoption part of the story better than the people who were there before," Marsland says.

The struggle to open source Solaris actually took much longer (it began about 1998 or so) because the opposition was stronger and the obstacles more numerous.

Marsland recalls his early work to free the code base from various legal encumbrances.

"That was a painful task," he says. "All the time we were being asked by different parts of the company, 'Why are you doing this? Why would you give our crown jewels away?'"

But to Marsland's way of thinking there are only two models that work in the software business:

"One is the Microsoft model where everyone dances to your tune because you have a monopoly. The other is the open source model where you get the world to help you," he says. "Those are the only two alternative business models with which we could address a multibillion-dollar market. We don't get the option of, 'Oh, I know, we'll just keep doing what we've been doing.'"

To him software is not about a particular code base -- what he describes as "a snapshot in time." It's about the people working on the code.

"The intrinsic value is in people's heads, and the fact that we have the people we have is what makes Solaris ours -- not protecting the code base with a particular license. Obviously if you stand still, everyone can steal your ideas, but if you're continually moving the bar, then everyone has to race to keep up. And we're the best people to keep up with us," he says.

"It's also the case that this is the way to build a community. It's about the conversation we get to have with developers and deployers about the deepest aspects of the operating system, how it works, where it's going. We're a bunch of engineers who like to build stuff people actually use, and this is like manna from heaven, because now we're actually talking to people who use it, directly, as opposed to that being filtered through surveys or other things. All of that is useful because you don't want to be influenced simply by people who are more vocal, but it's much healthier to be in a position where we're listening to both."

After just one year, the success of OpenSolaris -- 14,000 members, 29 user groups, 27 active projects, and more than 100 putbacks -- speaks for itself, he adds.

What's more, over 85 percent of the Fortune 500 now have the Solaris 10 OS in development or production.

Looking ahead, Marsland says the real challenge remains the same: making innovation and stability coexist.

"That's the art of the operating system here," he says.

In other words, how do you satisfy the needs of diverse customers -- some who want lots of cool new features to play with, and others who want only minimal changes -- no disruptions, please.

"We have virtualization technologies that allow those worlds to coexist more closely than they ever have before. They allow you to run complete operating system stacks side by side," he says.

That frees customers to experiment with the new while they continue to rely on the tried and true.

The old choice went something like this: "If I want to buy this box, I have to get this operating system. If I get this operating system, then I'll need to upgrade all the rest of my boxes if I want the whole environment to be manageable. That's a pain, because now I have to recertify all my applications, train all my people ... How am I going to deal with that? I guess I won't buy the box."

The new choice: "You can buy the box and run both the old software and new software on the same box. If you're ready to rush straight to the new world, go ahead. If you want one person or group to be able to play with the new software while everyone else continues to use the old environment, you can do that, too."

"With hardware virtualization, we can do that on both SPARC and x86 machines," Marsland says. "I like to speak of 2006 as being the year of hardware virtualization because the capabilities are now in these CPUs. We'll see people talking about the operating system plus its applications as a component stack that you can run alongside other component stacks as opposed to being the substrate on which everything else runs."

How the stacks interact, and the services they can provide to each other, is the new focus and Marsland sees "incredible opportunities and unbelievable amounts of work," but you get the feeling that's just the way he wants it.