calit2

Building Energy Efficiency into Networking

 

Gian Mario Maggio

Changing idle time into low-power idles creates
energy and cost savings in networking equipment.

Manufacturers have always juggled the tradeoff between performance and cost, striving to deliver high-performance products at reasonable prices. Now, for many manufacturers, including those who make networking equipment, there is a third variable in the equation: energy. 

During a
lecture at Calit2 last week, Nick Ilyadis, vice president and Enterprise CTO at Broadcom, described the company’s efforts to produce industry-leading performance features in a cost-saving, energy-saving framework.  “As we put more and more transistors on these devices and run at higher and higher rates, power has become a constraining function,” he said, adding that in data centers, the cost of power consumption could soon eclipse the cost of hardware.

Governments have begun mandating energy efficiency in different types of IT components.  Networking, which consumes a major percentage of all the energy used by computers, servers, and mobile devices, is no exception; Energy Star certification soon will include small networking equipment like the Ethernet switches Broadcom manufactures. “At the end of the day,” Ilyadis said,” lower energy usage means lower operating costs.”

The problem has been that even when traffic is not in transit, high-speed Ethernet systems continually send pulses of energy down the wire to keep both sides of the link synchronized. Laptop and desktop computers only utilize the link about 1 percent of the time on average, leading researchers to ponder the waste of energy when no information is being transmitted.

The solution is an Energy-Efficient Ethernet (802.3az) standard, one that can automatically reduce power consumption during periods of low link utilization, thus saving energy, money and the environment.

The 802.3az standard, which should be finalized later this year, calls for suspending energy on the wire by turning energy-wasting idling time into low-power idles. The Low Power Idle (LPI) approach powers down the Ethernet during inactive periods while maintaining the synchronization features that allow a rapid return to an active state.

It works like this: when a device recognizes that there is no data to send or receive, it basically “goes to sleep,” transitioning from active to low-power idling. This quiet period on the wire uses a negligible amount of energy. A periodic “refresher burst” maintains necessary synchronization, followed by more quiet time. When either side of the wire senses that information is waiting, an alert is sent that “wakes up” the electronics and re-acquires the channel. The process repeats as often as necessary.

In developing the new standard, Broadcom and other companies must ensure that no packets of information will be lost in the transition between idle and full power, and that transition time will be negligible.

Ilyadis expects the evolution in energy-efficient Ethernet to resemble that which occurred with cell phones. “In first-generation cell phones you had standby time of only hours; now it’s a matter of days,” he said. In the first generation of energy-efficient networking equipment, the physical interfaces will take advantage of the lack of traffic. But as design and architectures are refined, the electronics behind the physical layer will participate too.

“We’re also working the storage space, the server space, the PC space and the switch space,” he added. “These are all areas where we’re developing technologies and software to enable better use of energy.”

-- Anna Lynn Spitzer