With Internet traffic doubling every six months, networking is one of today's hottest markets. Linux has always played a key role in servers, but now the door is open for Linux to penetrate the heart of the Internet: the high-speed router. These devices, made popular by Cisco Systems, create the Internet by moving data from place to place.
Today, Cisco owns about 80% of the router market. Its products use a proprietary software stack called IOS to handle all of the routing functions. But, as the router market continues to grow rapidly, many new companies are attempting to dethrone Cisco. Rather than designing their products entirely from scratch, many of these vendors use a new model that relies on third-party hardware and software.
The catalyst for the new model is the network processor. This new device burst onto the scene a year ago and has already racked up more than 100 design wins with routers and other networking equipment. It replaces the custom silicon that Cisco and others painstakingly develop for each new product. Intel, Motorola, IBM, AMCC (through its recent purchase of MMC Networks), Lucent and Vitesse have all jumped into the network-processor market, and several start-ups are also developing products.
Internally, network processors are much like standard CPUs, but their architectures are optimized for handling the packet data that passes across the Internet. They bring the advantage of programmability to routers, allowing network equipment vendors to add features more rapidly to their systems, even in the face of quickly evolving standards. Buying standard silicon can eliminate the expensive and time-consuming design cycle for custom chips.
This model greatly reduces the barriers to entry that have helped Cisco repel invaders for the past several years. In this new world, the router market will begin to look like the PC market, with many new vendors jumping in building equipment using off-the-shelf silicon. This competition will reduce Cisco's dominance and provide router customers with more choices and lower prices.
To enable a more PC-like market, however, standard hardware must be joined by standard software. This is where Linux can play a role. Linux would not run on the network processor itself, which executes a fairly limited set of code at high speeds. But every router contains one or more control processors that oversee the system and perform any complicated or unusual tasks. For example, the control processor updates the routing table, an ever-changing map of the Internet, and translates any packets that are not in the standard protocols or formats.
For the control processor, many router vendors use internally developed software, such as Cisco's IOS, or a real-time operating system such as VxWorks. But some are moving toward Linux as a more open environment. For example, MMC recently joined with MontaVista Software to demonstrate an open-router platform running Hard Hat Linux. The platform was based entirely on standard silicon from MMC as well as open-software from MMC and MontaVista. Using these products as a starting point, an OEM could quickly develop and deploy a networking system that competes with Cisco's products. Even with the emergence of off-the-shelf hardware and software, the router market won't become as commoditized as the PC market. PCs are customized by their application software, whereas router vendors must develop all of the software that their equipment will use. Off-the-shelf products can greatly accelerate time-to-market, but OEMs will still add features and customize the product to gain an advantage in a competitive market. Most of this customization will be in the control-processor software. For this application, Linux has the advantages of a robust toolset and source-code availability.
As the networking market expands, more opportunities emerge for Linux. As the Internet grows, much of the new traffic is coming, not from corporate networks, but rather from DSL modems, cable modems and wireless connections. Network processors are starting to play a role in the aggregation points for these connections (known as DSLAMs, cable head-ends and cellular base stations, respectively). Where off-the-shelf hardware leads, open software is likely to follow.
Linux does not have an open path in this market; traditional RTOS vendors are also pushing their products. But, the networking community is certainly familiar with the cost and time-to-market benefits of Linux. Control-processor software can reach millions of lines of code, so software development becomes a critical issue in networking products. Smashing through $10 billion in annual revenue, the router market presents another sizable opportunity for Linux.