Pacific Connection(英語)

OSDL Attracts Linux Developers, Distributors, Bundlers, Users--and Linus Himself

Last April, Linus Torvalds traded his job at embedded chip maker Transmeta for one at the Oregon-based Open Source Development Labs. But while he changed employers, his actual job has remained largely the same. "It feels a bit strange to finally officially work on what I've been doing for the last twelve years, but with the upcoming 2.6.x release, it makes sense to be able to concentrate fully on Linux," Torvalds said in a statement. "OSDL is the perfect setting for vendor-independent and neutral Linux development."

Torvalds' "perfect setting" is an industry consortium of companies that have an interest in Linux: distributing and supporting it, bundling it on hardware, and using it at the IT level. OSDL was founded in 2000 by seven U.S. and Japanese companies (Computer Associates, Fujitsu, Hewlett-Packard, Hitachi, IBM, Intel, and NEC). Its main purpose is provide the equipment needed to test enterprise-class Linux applications, the kind that can actually run mid-size and large-size businesses. The hurdle confronting Linux as it takes a more central corporate role is that Linux developers tend to develop for uniprocessor machines. If Linux is going to take on Windows and Solaris as the OS of choice in the data center, developers need to get their hands on faster hardware. At the core of OSDL's operation is a server room full of test equipment.

But in an industry driven mostly by private enterprise, why does Linux need a consortium? Why not leave enterprise Linux development to companies already operating in this sphere? IBM, NEC, Hewlett-Packard, and Fujitsu must have plenty of screaming multi-processor machines laying around, and plenty of developers capable of programming them. IBM, in particular, has become famous for its Linux development, running on everything from embedded systems, to small servers, to clusters and mainframes. Why not leave the drive for Linux in the enterprise to IBM?

The answer is peculiar to the culture of open source development. When a single company puts resources behind Linux, the developer community gets suspicious. The company might amass a team of engineers who collectively devote thousands of hours pushing Linux forward. The company can argue that its motives are altruistic, that it's doing its good deed for the community, and that the community as a whole will benefit-no matter what hardware they are running.

It's a nice thought, but the community won't believe it. Fairly or not, the suspicion will be that the resulting branch of the Linux tree is tuned specifically for that company's hardware. That it looks like Solaris, or AIX, or some other Unix flavor. That the company is preaching community development, but furthering its own ends, as all companies do. The community will make that conclusion and turn its collective back on that branch of the Linux tree, letting it wilt.

"The hardware, software and services providers don't want to get locked into a single company driving the direction of the OS," says OSDL CEO Stuart Cohen. "They want multiple people providing the OS. They want the OS to be the same between the hardware and the application software layer, and they want to be able to take advantage of the [open source] value proposition." No single company can provide that certainty, he says. A conflict of interest-real or perceived-would always be present. "It's no different than what happened with the different Unix offerings from the different hardware companies," Ultimately, every company developed its own Unix flavor as a competitive advantage-in order to sell more hardware.

Tim Witham, OSDL's lab director since the beginning, thinks that the consortium does a kind of "impedance matching,"-in this case, balancing not an electrical circuit, but a cultural divide. "We impedance match between the open source developers and the large companies," Witham says. "There's a very big cultural difference between a large corporation, with its fiduciary responsibilities to its shareholders, and open source developers, who are doing stuff that matters to them right now." OSDL keeps the two sides talking, and gets the right people in touch. When Witham worked at Intel, he would hear complaints from open source developers who never got their calls returned. "The problem was that they asked an Itanium processor guy about a desktop chip set, so yes, he wouldn't get back. We understand that cultural difference. We can help developers deal with it."

Intel influence

That Witham came from Intel is perhaps no coincidence. OSDL is located on the fourth floor of a building in Beaverton, Oregon, seven miles west of Portland-which is also home to a large Intel campus (as well as shoemaker Nike). The connection between Linux and Intel chips has made Intel an early and enthusiastic supporter of OSDL.

"Linux maintainers tend to be focused, not surprisingly, on the kinds of systems they use themselves," namely desktop machines, says Andrew Wilson, strategic relations manager for Intel's telephony software division and chairman of the steering committee for OSDL's carrier grade Linux working group. "It's unlikely they've had access to 8-way and 16-way systems or to architectures delivering 'five 9's' [99.999 percent] availability." Wilson says that there's ample opportunity for Linux in the telecommunications industry's adoption of carrier grade Linux, which has been defined by such OSDL member companies as Alcatel, Ericsson, Nokia, and Cisco. Telecommunications has traditionally run on proprietary hardware running a proprietary OS. The prospect of substituting Intel CPUs running Linux represents a major potential market.

OSDL, which employs about 30 people, now includes hardware vendors like Sun, Alcatel, Cisco, Toshiba, Nokia and Ericsson, as well as Linux vendors Red Hat, and SuSE. OSDL is now trying to attract end-user companies, the most prominent of which, so far, is the food conglomerate Unilever.

OSDL has a second, smaller unit in Yokohama. "Linus told me that when you grow up in Finland you know you're going to have to learn English if you are going to exist in the technical world," Witham says. "But that's not true when you grow up in Japan." As a result, he says, Japanese developers are doing a lot of good things in Japan that are not immediately accessible to the English speaking world. The Yokohama facility "gives us a conduit" into Japanese Linux development.

The lab has also attracted Andrew Morton, lead maintainer of the Linux production kernel (just as Torvalds is the lead maintainer of the development kernel). He now works in association with the lab while continuing as principal engineer at Digeo, Inc. Neither Torvalds or Morton has actually moved to Oregon, but then, there's no real reason to. Witham can only remember one tester who actually walked through the door: a Texan in town during the Oregon ski season. Witham suspects that the snows probably beckoned. But every other tester has worked the way open source developers often do-by logging in remotely over the Net.

Testing code at OSDL

What developers primarily get from OSDL is access to the lab's test equipment: including storage to hold multiple terabyte files, and multiprocessor machines. Lab personnel serve largely to make the tests. They can tell you what's possible, untangle configuration problems, and otherwise help things run smoothly. They are mostly involved at the front end, not in evaluating the results-unless they're working with the lab on a joint project. "The testers are fairly sophisticated and know what they are looking for" Witham says.

Access to the lab is by application, but is otherwise free. The project itself must to be open source-with a license approved by OpenSource.org. And the code must usually be publicly accessible. The exception is experimental code that the tester is not confident enough to publish, but wants to test anyway. "In this case, they'll send me the code and I promise I won't send it to anybody else unless it turns out to be really good," says Witham. "The idea is they don't want to put it on a website when it's not really ready to go out."

OSDL test projects run the gamut: event logging and TCP/IP enhancements; tests of MySQL, Apache locking mechanisms, and Dynamic Probes (DProbes). There are testers developing more Linux benchmark tests. After the source code is submitted, the lab assigns an appropriate sized computer-anywhere from a two-way to a sixteen-way machine. The lab does an initial install of Linux, and the tester gets full root access. All the machines either have an internal power cycle board controlled through a console or a switchable power line. The idea is that, if need be, testers can actually perform a remote hard reboot. Machines are assigned for a fixed time appropriate to the project, which can usually be extended and often is.

"After the tests, we ask them to report back," says Witham. "But to be honest, most of them are not working on open source because they like writing reports. So we get comments like: 'Works good, got the data I need, thanks a lot.' That's the usual response." Most projects are ongoing. Testers get the results, go away for four or five months, and come back for more tests. That arrangement pleases Witham. "That's exactly the sort of resource we want to be," he says.

OSDL has helped streamline testing through its Linux Kernel Scalable Test Platform (STP), which lets developers perform rigorous testing of their code without getting bogged down in administering a test suite. Testers create, build and track software patches using OSDL's Patch Lifecycle Manger (PLM), then use the STP framework to run performance and scalability tests. (The tests are listed and described at http://www.osdl.org/stp/test_details/stp_test_descriptions.html). The advantage is that some of these tests can take considerable time to configure and run. "If you're a kernel developer, you don't want to spend the time doing that," Witham says.

Patches are submitted, compiled tested, and re-compiled more complexly. Testers can specify that the patch run against a specific test on a specified machine-say, an eight-processor workstation running a decision support system database performance test. Here, the system would do a clean OS install, as well as install the patch, database and the test. The test would be run, the results logged, and the tester sent results-all without intervention by OSDL personnel. "This represents another way of going about testing," says Witham. "You can either go in and have a dedicated machine to yourself, or interface with the STP. STP/PLM gives you less detailed control, and there are circumstances where you may need to do both. But if I had to give a machine to somebody, it would probably take them two and a half weeks before they made their first run. I can support six [STP sessions] a day, as opposed to one a month."

Sidebar: A conversation with Stuart Cohen, Chief Operating Officer of the Open Source Development Labs

CEO Stuart Cohen is the new guy at OSDL, hired last April to help bridge the gap between large corporate Linux users and the hands-on open source development culture. Cohen came to the lab from embedded board maker Radisys, but for 17 years, he was an IBMer, with positions in sales and marketing at the PC and networking divisions, including stints in China, Southeast Asia, and Europe.

Why were you brought in?

One big reason I came on board is to get the Lab talking about the awareness, the acceptance, and the acceleration of Linux. We will play a much more vocal role in doing this-talking about where Linux is going, how it's being used, and its value proposition.

Why does Linux need OSDL as an evangelist?

Linux has plenty of evangelists, but they come with their own respective business cards. IBM has their own Linux evangelists and they do a terrific job of promoting Linux as a value proposition-but it's in association with IBM goods and services. HP does the same, as does Oracle, Red Hat, SuSE, etc. But there is no one really talking about Linux from a vendor-neutral point of view. A lot of press, analysts, customers, value-added distributors and resellers are interested in getting a vendor-neutral perspective.

Would it be fair to say that you are trying to give it emotional parity with Windows and Microsoft?

Linux has tremendous emotional parity today. Of course, from a grand awareness and confidence standpoint, Linux is a lot less mature than Windows is-but that will change over time.

Do large and mid-size enterprises still have doubts about the viability of an open source operating system?

No, not at all. The IDC and Gartner research data says that Linux over the next five years is the only growing operating system on the server market. The reports predict that a combination of Windows and Linux will run in data centers around the world,. Linux is already clearly mainstream. Now, the question is: how broadly does Linux get accepted, how broadly will it be used, and how quickly does it get implemented in businesses of all sizes, in global businesses, consumer appliances, embedded equipment, servers, desktops, mainframes. There's a lot of different places that Linux can go, but its adoption rate will be different in each of those environments.

How close is version Linux 2.6 to becoming a viable enterprise OS?

Version 2.4 was very good. But 2.6 is a lot closer to the mark for big global deployment.

Where are some of the challenges?

Global service, global support. But they are growing with the need. The service and support industry was there when people began using Linux for Web servers, print servers, and file servers. As Linux is expanded to the edge of the network, services have expanded appropriately. As people now look at global deployment and tens of thousands of server platforms, the services and support organization will grow up around that. It's just a matter of the evolution of the requirements of the customer base.

How do you see Linux evolving in the US versus Japan and the rest of the world?

If you look at what Japan is doing from a consumer appliance standpoint, from an embedded standpoint, they have clearly made a big commitment to Linux. And there's a great deal of interest from a server, customer and government perspective. TurboLinux just joined OSDL. Miracle Linux is probably one of the more aggressive adopters of Linux. Certainly, if the economy were better, I think this would move even faster-the economics of Japan's market right now is probably hurting all technology.

Where do you think Linux is going to spread? Who are the next set of big customers?

The financial industry is clearly adopting Linux in a big way. Retail opportunities are very strong and growing rapidly. The new Version 2 security announcement that IBM and SuSE made at LinuxWorld will help governments. The consumer sector is growing rapidly, as is embedded. With our carrier grade initiative there's a lot of work going on in the telecommunication equipment manufacturers.

What's the significance of that initiative?

It's a place where that marketplace, the telecommunication equipment manufacturers like Nokia, Alcatel, Ericsson can join with IT vendors like MontaVista, Red Hat, SuSE, Intel, HP, IBM and the development community to work on requirements, standards and certification.

Microsoft has attempted to spread fear, uncertainty and doubt [FUD] about open source. Will that be a continuing problem?

I don't think they've done much. They've said that they see open source as one of their biggest competitors going forward. I don't think that's FUD-I think that's the reality of it. We would very much like to have Microsoft participate in OSDL, we'd like them to become a member. We think there's a lot of things that they have to offer to the open source community, and to the software stack above Linux, that would be valuable.

Not likely, though.

You never know. It wouldn't surprise me.

Sidebar-2 An Invitation to Software Design Readers...

As it has done in the United States, OSDL is establishing a Japanese customer advisory council comprised of larger companies. The aim is to develop Linux business requirements for the Japanese market. The group is slated to meet some time after the first of the year. If you are interested in this, or in other OSDL workgroups or services, please contact:

Shinji Takazawa takazawa@osdl.jp in Yokohama or

Nelson Pratt org">nbpratt@osdl.org in Beaverton, Oregon.

おすすめ記事

記事・ニュース一覧