Too Many Servers

Remember when the hot machine for running Linux was an Intel 486? Or a Pentium III? For most of us, single-chip Intel PCs are still the computers of choice. It's what we play on, what we work on, and what we develop on.

Virtual computer3

Remember when the hot machine for running Linux was an Intel 486? Or a Pentium III? For most of us, single-chip Intel PCs are still the computers of choice. It’s what we play on, what we work on, and what we develop on.

But it seems that more and more the standalone PC or server is not what most Linux users will be running on in the next few years. Oh, there’s still a movement toward making Linux a consumer desktop system. Indeed, there’s an excellent new Web site devoted to the subject (http://www.desktoplinux.com), and a December 2001 Linux.com poll showed that 25.6% of Linux users already think that Linux is a competitive consumer desktop operating system.

But since then, Loki, the one major Linux game company, went out of business. And even before Loki shut down, 17.4% of the people who took that same survey said they think that Linux will never be a big-time operating system for the masses. I’m with them. Sort of. You see, I think Linux will become a brand-name operating system for millions of business users. But they’re not going to be running Linux on desktop PCs, they’re going to be running applications on a Linux virtual machine (VM) running on a high-powered cluster or on a mainframe.

You could see signs of this coming change at this January’s LinuxWorld. Shirts and ties far outnumbered tie-dyed shirts. The big news of the show wasn’t that IBM was pushing Linux on the mainframe: they’ve been doing that for a while now. And it wasn’t their TV ads featuring Linux or their new slogan, “Linux is real business.” The big news, according to Bill Zeitler, head of IBM’s server group, is that out of the billion dollars IBM poured into Linux, “We’ve recouped most of it in the first year in sales of software and systems.” That’s not, “Oh, some day we’ll see a return,” or “Oh, it’s worth it because of good will,” that’s “IBM is making money from Linux today.

This isn’t just a retro-mainframe approach. International Data Corp. (IDC) expects Linux clustering, pioneered by TurboLinux, to become the dominant operating system in cluster environments, growing from $226 million in sales for the entire clustering industry in 2000 to $1.4 billion by 2005.

How is it being done? IBM is doing it by running Red Hat, SuSE, and TurboLinux as VMs on their S/390 mainframes and other “big iron” models. VMware is also in the picture with its ESX Server, optimized to run on IBM’s Intel-based eServer xSeries systems.

Compaq and Platform Computing are using clustering to follow a similar path on the Alpha. In addition, there are at least four open source VM projects, FreeVSD, Plex86, User-mode Linux (UML), and vserver. Commercial efforts in the same line come from SWsoft and Ensim.

You might be asking, though, “Why bother with virtual machines?” After all, it’s not as if your generic Linux box is going to run out of resources for ordinary server programs.

But there are several good reasons to go the VM route. One is security. While the Unix/Linux security model is strong compared to Microsoft’s, mainframes are much harder to crack. Even if a cracker does get root access to a single server running as a VM on the mainframe, he’s still no closer to getting into other Linux VM servers or the mainframe operating system.

Another advantage VMs have is stability. Half a dozen Linux VM servers can crash, but the other two dozen will still hum along normally. A standalone Intel-based Linux server simply can’t compete with that level of stability.

Finally, while Linux is getting better at symmetric multiprocessing (SMP), the IBM mainframe operating systems are much, much better. With the mainframe model, you simply don’t have to worry about SMP.

Of course, it’s not clear exactly how this will play out. While some are making a move from multiple servers to a few mainframes, Oracle is taking a different approach. In February, CEO Larry Ellison said that Oracle is replacing a trio of Unix servers that run most of its business applications with a cluster of Intel servers running Linux.

What’s clear, though, is that Linux’s biggest future role isn’t as the revolutionary operating system for PCs; it’s as the heart of corporate big-iron systems. There’s something ironic about that, but it’s the price of success. Linux is proving to be too good, too advanced for the desktop; instead its fate is to be the center of the enterprise. Who would have thought it?

Steven J. Vaughan-Nichols is a longtime Unix guru and technology writer. He can be reached at sjvn@vna1.com.

Comments are closed.