Building a network requires many individual components, each up to its own task and all running smoothly together
This is part of IEEE Spectrum's special report: Always On: Living in a Networked World.
The phenomenal growth of the communications industry is placing exorbitant demands on the components that run the networks. Routers, switches, and servers are constantly called upon to transmit more and more data at higher and higher bit-rates. Nor is the difficulty of their tasks declining, but the reverse. No longer simply packet shufflers, today's routers and switches must be able to examine and sometimes modify the packet payload for such operations as balancing server loads and collecting usage information for billing and network analysis.
But it is not enough for each component to operate at peak performance. All must work seamlessly together in systems that are constantly growing in complexity with the daily addition of nodes. To build networks that harmonize, designers use simulation to predict network behavior and laboratory and field tests to check out the real thing.
At the chip level, an ongoing development is the emergence of the network processor. Linda Geppert reports in "The New Chips on the Block" that these programmable devices, optimized to process packets, are replacing custom-designed, hard-wired circuits and general-purpose processors inside network components.
At the interface between the electronics and the transmission medium--optical fiber or air--one finds RF ICs that run at network frequencies. Now at 10 GHz for the highest-speed optical networks, frequencies will soon extend to 40 GHz and then beyond. To complicate matters further, cost factors are driving up the number of transistors on each chip, making far-from-simple design processes yet more intricate. In "RF Bridges to the Network," Geppert reports on performance criteria, materials choices, and design methodologies for these highly specialized chips.
The growing need for highly integrated communications ICs, particularly in the wireless arena, has equipment manufacturers looking toward another emerging technology: microelectromechanical systems, or MEMS. These devices are generally made with the same processes used for ICs, so they can be integrated onto the same chip with other circuitry. MEMS also promise power savings over today's surface acoustic-wave (SAW) filters and solid-state switches because they can be used to make passive components like switches and filters. In "Large Jobs for Little Devices," Stephen Cass describes recent advances in the development of MEMS for communications.
Even before all of the network elements are chosen, network architects begin to assess the factors that will affect system performance. To do that, they call upon network simulators, reports Gadi Kaplan in "Simulating Networks." These tools allow them to examine how their network behaves under varying conditions of transaction rates, routing protocols, and applications. A new technique called hybrid simulation combines analytical and discrete-event techniques to speed up the process while giving the designer the required information.
Still, in the end, real data has to run on real hardware. And with communications technology outrunning the ability of test equipment to measure performance directly, network architects and component builders are taking their products into the laboratory to see how reliably they work with other network elements. Field trials, which often involve real customer sites and many kilometers of buried cable, are also essential steps in verifying the most advanced systems, those that run at 10 and 40 Gb/s, says Elizabeth A. Bretz in "Network Test Starts in the Lab."