With its Irix OS on life support, Silicon Graphics Inc. turns to Linux to save the company.
Is Silicon Graphics, Inc. a Linux company? Many old-school Unix vendors — Sun, HP, IBM, even SCO — have embraced Linux, with varying degrees of sincerity.
But none of these companies has launched an in-house development effort on a par with SGI, and none has pledged to deliver as much technology to the Linux community. Though SGI still has its own version of Unix (Irix) to support, the company’s dire financial straits have given its Linux endorsement an urgency that is not matched by other proprietary Unix vendors.
SGI’s business strategy has taken some twists and turns in the last few years. The company has lost money in nine of its past 11 quarters. Sales of its high-performance Unix-based workstations that once outshone its competitors in data- and graphics-intensive environments — such as those that produced the sophisticated special effects in movies like Star Wars and Jurassic Park — have slowly but surely been eaten away over the years by low-cost, high-performance PCs from the likes of Compaq, Dell, and Gateway.
In early 1999, SGI began selling Windows NT machines, which were to be the heart of a new line of business for the struggling company — only to backpedal in August, acknowledging that the effort was a failure and pledging to phase out its NT business in favor of Linux.
At the time, SGI’s then-CEO Rick Belluzzo spelled out the company’s Linux strategy forcefully, stating that graphics workstations running Linux — optimized to manage multimedia tasks and large data files — were ideally suited for high-end Internet applications. Two weeks later, Belluzzo,
having just orchestrated a massive reorganization of the company and delivered its first quarter of profits in quite some time, jumped ship to head up Microsoft’s Internet division.
Return of the Engineers
Company insiders say that SGI’s refocus on Linux has largely been driven by engineers inside the company, who have found themselves pushing a management team struggling to improve profits. According to engineers and executives at the company, since Belluzzo was replaced by SGI marketing whiz Robert Bishop — a relative unknown in the industry — the company’s focus on Linux has continued to intensify.
“The change to Linux is being driven by technical people,” says open source advocate Eric Raymond, who points out that SGI has traditionally been an engineering-driven company.
SGI Senior Vice President of Computer Systems John “Beau” Vrolyk predicts that over the next five years, “We’re going to see the decomposition of operating systems as we’ve known them.”
“In the past,” he continues, “an OS offering from a proprietary vendor included many things beyond the core OS. Now, with Linux being free, everything will start being unbundled. The customer will be able to buy what they need, rather than having a gigantic stack of CDs delivered regardless of their usage pattern. The core OS, which must remain entirely open to be successful, will stay free, while the utilities and other components will be priced to market forces.”
The rate at which SGI customers are jumping on Linux, however, has come as a surprise, according to Director of Strategic Technologies David McAllister.
“What they develop on is what they deliver for,” he notes. “As people are developing for Linux, application availability starts moving up. Linux blows the market wide open by providing a Unix-like operating system on extremely low-cost Intel-based hardware.”
Engineers and management at SGI seem to be in agreement. “You can’t touch a technology at SGI that is not being considered for Linux,” says McAllister.
SGI’s long-term strategy is to gradually make Linux as important as its high-performance Irix operating system. But for some industry watchers, the strategy lacks focus. Rob Enderle, analyst at Giga Information Group, Santa Clara, CA, says, “Irix was dead when SGI announced their Windows NT strategy.” Enderle believes SGI has taken the wrong path. “Linux is entering at the lower end of the market on low-cost servers,” adding, “SGI now has a brand conflict. It’s as if Cadillac started selling the Geo. It’s the wrong platform now.”
Giga’s Enderle acknowledges that eventually Linux will “grow up” and be scaled up for high-end applications, but for the time being it is not suited to SGI’s traditional data-intensive environments.
“SGI is on the other end of ‘free,’” he says. “They will do great good for the Linux community with their technology contributions, and certainly the technology is interesting. But high-end cluster buying cycles are very long. Linux doesn’t fit in there yet.” SGI is trying to prove otherwise.
Giving to the Community
|Snatched by VA: Jeremy Allison|
SGI’s contributions to Linux technology and the Linux community at large have been steady and substantial, and somewhat unheralded. From SGI engineers’ contributions to the Linux kernel itself, to two full-time positions devoted to developing the Samba Windows file and print software for Unix, to XFS, the world-class SGI journaling filesystem, and a range of other technologies donated to the Linux community, SGI is making a dramatic impact on the Linux code base. In some cases, however, the company is still struggling to “clean the code” so it can be handed over to the Linux community, an effort that company insiders admit is formidable. “They have to make sure whatever they release is not owned by somebody else,” says Samba developer and former SGI employee Jeremy Allison.
SGI’s credibility in the open source community suffered a setback this fall, when Allison, SGI’s most prominent open source developer, announced he was leaving SGI to work for VA Linux Systems, but Eric Raymond says the fact remains that SGI has been a community player and continues to make major contributions to Linux technology. “We’ll see the impact of its technology contributions to Linux in the next four months,” he predicts.
The End of Irix?
SGI first recognized the bottom-line benefits in developing technology for Linux and all of the new markets it represented back in the summer of 1997, when what became known as the Group of Eight met to aggressively pursue the promise of Linux.
In July 1998, SGI’s Group of Eight was headed by Kurt Akeley, an SGI founder. He was joined by six principal scientists and engineers, and David McAllister, SGI’s director of strategic technologies.
“We were looking around to find out what other things were on the radar scope,” says McAllister, explaining that his job is to figure out where the world is going to be two years down the road, and help SGI get there. “At the time, Linux was still relatively unknown,” McAllister remembers. “It wasn’t until after Oracle endorsed it, that it suddenly caught on.”
By that time, SGI’s Linux plans were well underway. The company recognized that Linux was a stable version of Unix that was dominant in the Internet arena, a market segment in which SGI needed to play a greater role. SGI’s original focus was in the small-server category — which included Internet, education, and research-and-development environments. “What was surprising was the rapid adoption of Linux by commercial software players,” McAllister says.
In early fall 1998, the Group of Eight brought its recommendations to SGI’s top executives — Rick Belluzzo and members of the board — and received the go signal to begin implementing a major shift to Linux. “We identified where Linux overlapped with SGI’s markets, and where it complemented them. We owned the high end of every marketplace we touched,but needed to participate in the small-server marketplace, which fits very well with Linux,” McAllister says.
There were surprises almost immediately. “It’s a groundswell movement,” McAllister says. “By stating that Linux was one of our chosen operating-system directions, we found out that a lot of our engineers were already running Linux at home. They were already up to speed on it.”
In April 1999, SGI formed an internal organization known as the Open Source Review Board. Its mission is to look at the company’s technology offerings and determine which ones make sense to offer as open source. The group is made up of marketing, engineering, support, and legal staff, as well as SGI executives. By August 1999, just four months after its creation, the review board found itself playing a much more prominent role.
SGI’s plan is to gradually bring most of Irix’s features to Linux. The company is looking to hire Linux programmers who are comfortable in the open source world. “There is a cultural adjustment that is being made moving to open source from proprietary,” Allison says. “SGI is looking for people who are already in that world.”
“Irix is dead,” Raymond says. “You’ll hear different things because they have to support their customers. But in 1999, the OS business is a stupid one to be in. It’s an issue for everyone, not just SGI. Old-line Unix vendors recognize this.”
But Allison insists that Irix will remain vital in the supercomputing world for some time. “SGI will never turn Irix into Linux,” he says. At the same time, he believes the company recognizes that the very high end of the market is less profitable. “Irix will become a supercomputer operating system. Linux will gradually become more important.”
A Merger with VA?
|Eye on SGI? VA’s Larry Augustin|
The landscape is getting interesting. In mid-October, SGI announced that, along with VA Linux Systems and O’Reilly & Associates, it would jointly support a distribution of Debian Linux. The move was largely viewed as an effort to ensure that no single distribution — Red Hat’s, for example — would overpower the market. Engineers at SGI, also an investor in VA (of which Raymond is a board member), have worked closely with VA engineers over the months, and the relationship between the two companies may get even closer. Industry observers say that if VA’s IPO is a success and if CEO Larry Augustin gets his wish, he may very well just buy SGI outright.
“It’s a perfect fit,” says a source close to VA Linux Systems who asked to remain unnamed. “The engineering teams know each other well, and work together well. It makes a lot of sense.”
At its present rate of about $10 a share, SGI — with its 8,000 employees — has a market value of just under $2 billion, but in a market where a tiny company (150 employees) like Red Hat Software can be valued at over $4 billion, the idea that a post-IPO VA Linux Systems could acquire SGI is actually more than a remote possibility.
The analysts aren’t completely cold on the idea of a merger. Stacey Quandt, also an analyst at Giga, says the possible purchase of SGI by VA could be advantageous. “Marrying of the two companies would be very compelling,” Quandt says. “Larry Augustin may also want to acquire SGI because his company is a smaller player in the Linux market than the likes of IBM and Compaq. This would expand VA’s mindshare and market share.”
Linux allows SGI to enter a number of new markets, according to SGI’s Vrolyk. “Up to the introduction of the SGI 1400L [workstation], our first Linux product, SGI didn’t have a low-end 32-bit product with which to compete with Sun and HP,” he says, referring to the fact that while Linux is a 64-bit OS, it is most commonly run on 32-bit Intel machines.
Irix, SGI’s 64-bit OS that scales to 256 processors, is much larger than necessary for the kinds of applications that people want to run on Intel systems, according to Vrolyk. SGI projects that Linux, which right now is optimized for single- and dual-processor machines, will grow most quickly in the traditional “early adopter” sites. These include many of SGI’s traditional customers.
“Research labs, universities, and advanced development organizations are all well down the path to adopting Linux,” Vrolyk observes. “They will show the way, as always, to the enterprise computing customers who invariably adopt technology more slowly.”
While many companies, including Sun Microsystems, have been paying lip service to Linux, it is apparent that SGI is investing far more resources and effort in contributions — technology and other — to the Linux community. According to Vrolyk, this is because SGI built only high-end 64-bit systems prior to its introduction of the SGI 1400L and doesn’t face the internal competition that other companies do.
“In particular, Sun faces the complete destruction of its closed, proprietary low-end systems business by the completely open Linux operating system,” he adds. “Because the success of Linux is strategic to SGI for its new low-end products, SGI is pouring technology into the open source community and fully supporting the use of the GPL as the correct contract for such donations.”
Shifts in the market are expected to occur when 64-bit processors like Itanium (formerly known as Merced) are on the market, and Linux is powered for these processors. SGI’s Vrolyk says that market segmentation is determined by the application’s software as well as the performance of the system, however, and isn’t linked nearly as strongly to the number of bits of the CPU, “nor is it linked specifically to the operating system,” he notes.
Vrolyk remarks that in 1999 Linux has picked up support for a massive number of applications and has grown along with the growth of the Internet. “The key advantage Linux will have, as Intel systems move from 32 to 64 bits, is that applications from the 32-bit systems will simply run
on the new 64-bit systems without change. The major shift in this market will occur as low-cost Linux systems obsolete closed proprietary architectures like Solaris,” he predicts.
To date, SGI has contributed a number of technologies to the Linux community and is working on others. At the kernel level, working closely with Linus Torvalds, SGI engineers contributed a kernel debugger known as KDB.
Prior to KDB, Linux developers used a technique known as “wolf fencing” to identify and fix bugs in the Linux kernel. The term goes back to the early days of programming when debuggers didn’t exist. The metaphor refers to a practice on European farms, where sheep were kept in separate, small pens secured by fences. Wolves would have to leap over the fence to get at them, so farmers could trap wolves by luring them with a small section of sheep rather than exposing their entire flocks. Like isolating a small group of sheep from a predator, programmers would debug the kernel by narrowing down possible bugs by printing up statements and determining where the system died, scouring through final print statements before a crash.
KDB was released for Linux in the spring of 1999. In February 1999, SGI embarked on its GLX graphic library extension for Linux. It is a reference port for how to write efficient Open GL drivers, funded by Red Hat and SGI with engineering work being done by Precision Insight, Inc. The technology has been demoed at Linux trade shows and will be available for developers by year’s end.
Perhaps SGI’s biggest contribution to Linux will be its XFS journaling filesystem. Journaling filesystems allow administrators to recover systems without loss of data in the event of a power failure or other mishap. Currently system crashes can be checked with the fsck filesystem-checking utility, but fsck can take a long time — weeks sometimes — to accomplish this, especially when dealing with huge amounts of data.
SGI is now in the process of doing the “legal cleaning” of the XFS code to make it open source and porting the code to Linux. The first set of code was released in August, and the company is targeting a more complete release in early 2000.
Open source projects sponsored by SGI
In the last year SGI has contributed to a number of open source projects and released several pieces of code — from nifty patches to major pieces of OS infrastructure — itself.
Bigmem– A Linux patch that lets Linux systems support 3.8 GB of physical
memory. Released in 1999.
CRCalc — A very precise calculator. Released in August 1999.
GLX — OpenGL extensions to X. Expected by 2000.
Jessie — A Java-based debugger. Released in August, 1999.
lkcd — A Linux utility to gather information on kernel crashes. Released in 1999.
Linux/MIPS — A port of Linux to SGI’s MIPS architecture.
Lockmeter — A Linux kernel lock-metering patch. Released in 1999.
OpenVault — Storage-management software for removable media. Released in August 1999.
Samba for IRIX — In 1998 SGI became one of the first Unix vendors to ship Samba with its proprietary Unix.
SGI KDB — A kernel debugger. Released in spring 1999.
SGI Linux — SGI’s Linux distribution. Based on Red Hat 6.0. Released in August 1999.
XFS — A journaling filesystem. SGI expects to release full source code in early 2000.
Among the areas in which SGI is intensifying its Linux efforts are high-end systems such as the 128-processor Beowulf cluster it is now shipping to the Ohio Supercomputing Center.
“There’s been more and more news around Linux-based clusters,” McAllister says, and an engineering group inside SGI is doing nothing but Linux-based cluster work. A primary benefit of using Linux at the heart of these high-performance systems is cost.
“If you have lots of machines and need an operating system for every one — 32 machines at four processors apiece, for example — 32 copies of Linux cost you the same as one copy of Linux,” McAllister says. “Thirty-two copies of NT costs you 32 copies of NT.”
The company is working with a number of customers to develop clusters of the SGI 1400L Linux workstations. SGI sees Linux clusters as the next logical step in a specific class of multi-CPU architecture. VP Vrolyk says many supercomputer applications run quite successfully on Linux clusters, effectively removing the requirement for “cluster-in-a-box” solutions like the IBM SP2.
How will Linux solutions and technologies potentially contribute to SGI’s profitability?
“Linux will grow, over time, to become a major part of SGI’s business,” Vrolyk says. “At this time, however, it is dwarfed by the success of our MIPS/ Irix based products. As the entire marketplace moves to truly open systems, leaving closed architectures like Solaris behind, SGI will be perfectly positioned to lead that transition and grow with the market.”
Over the next year, Vrolyk predicts, “Linux will extend its lead in the Internet market to well over 50 percent marketshare. The growth rate in Europe, where Linux is already the number one OS for Web serving, is a good indicator. Other markets, such as CAD [Computer-Aided Design], EDA [Electronic Design Automation], and research will also develop rapidly. Growth rates in excess of 150 percent are certainly possible.”
What’s more, an explosion is expected to occur once Intel’s new 64-bit Itanium chip is on the market. SGI’s participation in the Trillian project, a consortium made up of a number of companies including HP, IBM, VA Linux Systems, SGI, and Cygnus Solutions, will ensure that Linux is tuned for Itanium the minute it hits the market. “With 64-bit [technology], the sky’s the limit,” McAllister says. There’s going to be some blurring of the more traditional operating systems and the more traditional processors. They’re going to have competition they haven’t seen before.”
The most surprising area of growth for Linux in the past year has been in the “bioinformatics” market, says McAllister. “They have ported more code to Linux than any other single area.” This market includes genetic and medical research and was an area that SGI predicted would be a big one for Linux, “though we didn’t think it would happen this fast,” McAllister says. That’s because the bioinformatics industry “is a huge multinational industry in which sharing of information is everything. Likewise, open source is a multinational solution where sharing of information is everything.”
SGI has its work cut out for it. It’s got a new CEO, a new operating system, and an entirely new business model. It is laying off staff and making its second run at selling Intel-based systems. But if any one technology is better equipped than Linux to be a catalyst in turning around this former darling of Silicon Valley, there are some engineers in Mountain View who would like to know what it is.
Wendy Goldman Rohm is Editor at Large for Linux Magazine. She is the author of The Microsoft File: The Secret Case Against Bill Gates and Under the Radar, written with Red Hat CEO Bob Young. She can be reached at firstname.lastname@example.org.