Decisions involving the adoption of a new operating system depend on a variety of factors, including installed base, robustness, and portability.

MIKE DOWNING
Senior Editor

Is Cisco trying to move away from its proprietary IOS platform? Is Linux robust enough to play in the high-end router space? And how easily does NetBSD port to Linux? The answers to these questions, as provided by router vendors Pluris, Hyperchip, Juniper, and Cisco, depend largely on past decisions.

According to Bora Akyol, a software architect and operating system (OS) specialist at Pluris Corp. (Cupertino, CA), his company uses a commercial system that it purchased from a well-known real-time operating system (RTOS) vendor. "You can guess for yourself," says Akyol. "When we started about three years ago, the freely available operating systems weren't quite what we wanted. So, we decided to go with a commercial vendor."

The climate has changed since then. If he had to do it all over again, Akyol says he would strongly consider using a freely available and open-source OS. "This could be anything from NetBSD or FreeBSD or Open BSD to Linux RT, and even some esoteric stuff like Darvin, which is from Apple," he says. "Open-source software is maturing rapidly, and it's nice to have access to the source code when you're troubleshooting."

Speaking of NetBSD, Richard Nor man, president and chief technical officer of Hyperchip Corp. (Montreal), says his company is currently using NetBSD on its routers. In contrast to FreeBSD, NetBSD was designed specifically for networking. "Our path is very similar to Juniper's," says Norman, "but instead of starting with the FreeBSD, we started with the NetBSD OS. From there, we made enhancements to the OS to fit the requirements of our routers."

John Stewart, marketing engineer for Juniper Networks, confirmed that Juniper began with FreeBSD and modified that product to fit its own requirements. The product is known as Junos. "We had to modify FreeBSD a fair amount," he says. "We took the networking part in the Free BSD software, threw it away, and replaced it with our own specialized software. That way, we don't have to worry about file systems and process management and all the operating features that the OS community is better at doing. We focus on adding our value to the networking part."

Stewart likes the development communities that stand behind FreeBSD, NetBSD, and Linux. He likes FreeBSD because he believes that very intelligent people are active in the community and they make sure the software includes the latest updates. "For example, several security tweaks related to TCP have been added in recent years," he says. "And FreeBSD is one of the first places where such things become available. Therefore, by staying in sync with FreeBSD, we're able to keep up with the latest improvements in software development. The combination gives us robust software that exists in a system that has a clean separation between the control and data planes."

Peter Long, director of marketing for the Cisco's IOS Technologies Div., says Cisco's OS model began 15 years ago. "Unix was just coming out of the universities," he says. "You had to buy a license from AT&T, and it was really expensive. So the guys who started Cisco realized they needed software that was fast and nimble and enabled them to build a set of intelligent services on top of their routing hardware platform."

In writing IOS, Cisco attempted to mimic Unix-like functions, including a scheduler to manage processes, timer services, and processes to manage memory and hardware resources of the box. "It has evolved dramatically since then," says Long. "And is now a significant differentiator between ourselves and our competitors."

The three major usage trends in router operating systems include proprietary software, commercial royalty-based software, and open-source non-Linux software. Then there's Linux.

Akyol is a big fan of Linux. "Linux gives us lots of options," he says. "For example, you can run a Unix-based system on the control plane for its memory protection and for the tools and source codes that are available. Then you can run an RTOS on the line cards and a driver that will basically transmit information from the control cards to the line cards. These are some of the alternatives that I see when I look at Linux. I also see a lot of people, especially in the networked- appliance space, using Linux in their development. This includes storage."

At the same time, Akyol believes the cost model for Linux is not mature yet. "I recently talked to one Linux RTOS vendor," he says, "and their prices were exorbitant." Akyol says that, although the Linux community does a great job advancing the technology while contributing to the open-source community at the same time, not all of the companies understand the cost model. "They have focused so much on small companies that they do not understand development within larger companies. In addition, because the OS itself is essentially free, these companies try to make money from support, but most of us don't need that much support."

Akyol says that if Pluris were to make the decision to jump to an open-source OS, it would be because they would want to own the code. "We would develop the code from there," he says, "as opposed to buying package support from a vendor."

Like Akyol, Hyperchip's Norman also recognizes the potential strength of Linux. He sees a number of vendors, including Hyperchip, looking at Linux. "I expect that many vendors will move to Linux because the progress on Linux is faster than on any other OS," he says. "But Linux-in my opinion-is not yet stable enough to compete in the routing space."

That's not to say that Linux isn't catching up. "Linux is still behind in terms of stability and robustness, and routers must be very robust," says Norman. He says that although Linux is way ahead of Windows in terms of robustness, it's not quite as stable yet as some of the versions of Unix that are available. "If we were starting two years from now, we would probably start with Linux," says Norman. "But starting two years ago-as we did-NetBSD was the right call."

So although Norman believes Linux has not yet arrived, Hyperchip continues to think about making a shift. "Code written for the Unix family ports easily," he says. "Therefore, if Linux gets there and is stable enough, we could take our modifications to NetBSD, put them into Linux, and get them accepted as a standard." Because Linux modifications rapidly become part of the standard, Norman says, the standard catches up very quickly with reality. "With other operating systems, modifications don't get automatically rolled back into the original. That's why we believe Linux will catch up very rapidly."

Juniper's Stewart says the strength of its Junos OS is its flexibility, which makes life easier for suppliers. "Be cause we start from a FreeBSD base, our suppliers don't have to provide exotic parts," he says. "The main requirement for a supplier is to provide us with an Intel-based PC on the board that can run FreeBSD. And that requirement is something most vendors have as part of their stock regression tests. Does it boot Windows? Does it boot Linux? Does it boot FreeBSD? So it's actually very good in terms of our requirements for those component vendors, because they don't need to do anything particularly wacky to meet our needs."

Peter Long says Cisco is not in any hurry to move to something commercial or "free." "Linux or FreeBSD or NetBSD basically provides a very small piece of a routing OS," he says. "That's not where we're spending a lot of time these days. We're spending more time developing MPLS (Multi protocol Label Switching), security, and solutions for interfacing with wire-rate capabilities like OC-192 and multiple-gigabit Ethernet platforms."

That said, Long concedes that if Cisco were a startup today, it would likely go out and get a NetBSD or a FreeBSD and work with that. "It would provide the basics to get started," he says. "You'd have a timer, a memory block, and instructions for running processes. But we already have that, so why switch?"

Long also stressed that contrary to opinions he hears voiced in the industry, IOS is not "big and slow." "Quite the contrary," he says. "Some of the folks in the industry tend to get confused. It's true that we offer a great deal of functionality on top of the OS. But the OS itself is very nimble."

Making the decision to move to another router OS is not an easy decision. There are a number of difficulties to consider, including how such a switch would affect the installed base and how such a switch would impact time-to-market decisions.

Then there are aspects that might not hit you right away. For example, Akyol says Pluris is primarily interested in the development tools that come with the OS, not the OS itself. "We have our own networking stack and we have our own routing protocol codes," he says. "So we don't really reuse anything that the vendor gives us in that regard. Instead, we're looking for a strong debugging environment and the closeness of the kernel level support for the networking stack to the BSD 4.4 environment. That's the gold standard of network programming."

Then there are the considerations of speed and robustness. "You must have speed or customers won't even talk to you," says Norman. "But once you reach a certain speed threshold, you could be twice as fast and customers wouldn't care. Either you have enough or you don't. Once you hit the speed threshold, robustness becomes critical. And that is the primary reason we haven't moved to Linux yet. It simply isn't ready. Don't get me wrong, compared to Microsoft Windows, Linux is rock solid. But compared to Unix, it's not as good."

Scalability is another challenge of an OS. "Customers want to know they'll be able to move forward," says Norman. "They like to see a road map. Any router OS needs to provide that road map."

Standards are also a concern, as are sourcing issues. "Our customers want to make sure that the standards bodies specify new protocols or extensions to existing protocols," says Juniper's Stewart. "That's why we tend to stay away from the proprietary models. Those bring in all kinds of problems, like single-source manufacturing concerns on the hardware side."

So what does the future hold for these four companies? Akyol says that although Pluris is perfectly happy with what they have, they are also currently considering other options. "We would like to add features eventually," he says. "But any decision to shift to a new OS would be based on the availability of additional fault tolerance and reliability in the OS. If we could get an OS that had improved memory protection, good isolation between processes, and fault tolerance support for replication of data structures in the kernel layer as opposed to the process layer, we would definitely be interested."

"We are definitely looking at what else is out there," he continues. "However, once we make a decision, we're going to stick with what we choose, because there's so much work associated with developing the OS."

Stewart says that Juniper will likely continue to follow its current model: "My opinion is that our current model will live a long life. I think we're on the right side of the build-versus-buy equation in that we build the parts that involve our core competency and buy the parts that are not. Other companies don't have that flexibility because of legacy reasons and installed base. We can continue to leverage the freely available OS while we add value through our core competency. That way, we don't have to spend a lot of time working with the kernel. We think our current model scales very well going forward to spend time on the things that meet users requirements."

Norman sees NetBSD as the near future and Linux as the extended future of Hyperchip. "Porting from NetBSD to Linux and back again is a reasonable project. I say this because the same mindset that wrote NetBSD wrote Linux, so if it turns out that a year from now Linux becomes more stable than NetBSD, we could move over in a matter of a couple of months. It's a nice position to be in. Today, however, when we project the stability and robustness of Linux, it looks like it will catch up in about two years."

And finally, contrary to rumors circulating throughout the industry, Cisco is not interested in moving to another OS. "We let our customers lead us," says Long. "They want things like scalability. When IOS was first developed, it was for a platform that had three interfaces on it. As we now look at some of the platforms we're delivering, such as the DSL aggregator product, we can support 30,000 to 40,000 virtual interfaces with IOS. So we've taken the customer requirement, which is scalability and in some cases re-architected elements of the IOS kernel to make it more scalable. We've taken advantage of hardware acceleration and hardware performance to make the performance angle work. We've separated routing and switching and forwarding to take advantage of those platforms. If it were the same IOS from 15 years ago, you could make those claims-and we might be looking for another OS-but his is not your father's IOS. We've come a long way and we're not focused on making a switch in operating systems."

It's clear that these four router OEMs plan to stick with their current operating systems for the foreseeable future, at least until Linux becomes more robust.

To get the quickest response to your question, contact the appropriate person listed below.
Reprints, Bulletin Board
Shirley Lessard
Tel: 603-891-9282
E-mail: shirleyl@pennwell.com

Subscriptions, back issues
Michelle Blake
Tel: 603-891-9360
E-mail: michellb@pennwell.com

New Products, Calendar
Suretta Williams
Tel: 603-891-9362
E-mail: surettaw@pennwell.com

News and features
Stephen M. Hardy
Tel: 603-891-9454
E-mail: stephenh@pennwell.com

Media kit, editorial calendar
Hilary Lloyd
Tel: 603-891-9218
E-mail: hilaryl@pennwell.com

Paid subscriptions
Omeda Customer Service
Tel: 847-559-7520
E-mail: lw@omeda.com