Here comes energy efficient Ethernet

When most people think of the Energy Star program, the products that probably come to mind include lighting, home appliances, and consumer electronics. But as government policy makers accelerate and broaden efforts to save energy, several industries are paying closer attention, reasoning it is better come up with initiatives of their own before mandated to do so.

That's the case with the 802.3az Energy Efficient Ethernet (EEE) standard from the Institute of Electrical and Electronic Engineers (IEEE). As its name implies, it aims to reduce energy on wired Ethernet connections. It was adopted by the IEEE last September. (See “The Road to the EEE Standard,” next page.)

Ethernet, of course, is the dominant wire-line technology for communications in computer networks. It is also being considered for use in access networks and even for long-haul links.

In a nutshell, the IEEE scheme saves energy by blasting data over Ethernet lines, then putting the network controllers to sleep until they're needed again. More specifically, the .az portion takes aim at the four different standardized rates used for unshielded twisted-pair (UTP) wiring which account for the vast majority of present day Ethernet links: 10 Mbits/sec (1-BASE-T), 100 Mb its/sec (100BASE-TX), 1 Gbit/sec (1000BASE-T, and 10 Gbits/sec (10GBASE-T).

Estimates are that the 802.3az EEE standard could save $400 million a year just in the U.S. alone. This figure is certain to climb as Ethernet data rates rise in that it takes more power to hit faster data rates. For example, a transceiver chip for the 1000BASE-T Ethernet physical layer (PHY) typically consumes about 0.5 W of power while a 10GBASE-T device consumes about 5 W.

Michael Bennett at the Lawrence Berkeley National Laboratory (LBNL) and chairman of the IEEE P802.3az Task Force says the new standard could account for energy savings ranging from 1.73 to 2.60 TWh/year just in U.S. residential equipment alone. That translates into $139 to $208 million a year in energy cost savings. For commercial data center equipment like servers, storage systems, switchers, routers, and so forth, energy savings could run anywhere between 1.47 to 2.21 TWh/year just in the U.S. That translates into a $118 to $177 million-a-year savings.

The official term for the energy saving technique proscribed by the IEEE P802.3az Task Force is Low Power Idle (LPI). There is a general consensus that it can be challenging to quickly turn on a dormant network card, as LPI prescribes. But it is much easier than an alternative method the task force evaluated called Adaptive Link Rate (ALR). ALR uses a two-way media-access control (MAC) frame handshake. This can be implemented in either the device driver or within the Ethernet controller.

The problem: This concept is time consuming when switching between Ethernet speeds (i.e. from higher to lower data rates and vice versa). Changing the link rate means dropping the link rate then re-establishing it, which can take several microseconds, too slow for modern networks.

The LPI concept was hatched at Intel. “You're better off sending data faster and getting to sleep quicker, which allows you to save more power over the long haul,” says Robert Hayes, Intel strategic planner for networking products. Most Ethernet data traffic comes in bursts and is thus suited for the LPI approach, he explains.

Link utilization, packet transmission time, and the distribution of packet intervals all determine the energy efficiency of the EEE concept. Task force members believe that it is reasonable to assume that EEE overhead consumes the same amount of power as packet transmission. Many studies have found that network link utilization, especially at the edge, is generally low, driven by the need for high-speed bulk data transfer or demand.

Continue on next page

One aspect of EEE performance improvement is through a technique called packet coalescing. This involves the use of a first-in first-out (FIFO) memory queue in the Ethernet interface (in the host network interface controller or NIC card and switch or router line card). The FIFO collects, or coalesces, multiple packets before sending them on a link as a burst of back-to-back packets. Packet coalescing is already used in many high-speed Ethernet interfaces (mostly on the receive side) to reduce CPU overhead for packet processing.

Coalescing is based on packet count and/or time from the first packet's arrival. It should, in most cases, require no additional buffer memory. It is a function of when packets that would otherwise be queued in system memory get released to the NIC for transmission.

Silicon interest

Chips that comply with the IEEE 8092.3az standard have been announced and some are in field testing. Even before the IEEE 802.3az EEE standard was ratified, semiconductor companies displayed a lot of interest in producing energy efficient Ethernet NIC interface chips. One of the first companies to implement an Ethernet energy efficient feature is D-Link Corp., which manufactures eight-port 10/100/1000 desktop switches using Ethernet PHY ICs from Vitesse Semiconductor Corp. Its D-Link Green technology claims power savings up to 84% when connected devices are powered down. These power savings result from detecting the links status and the link-cable's length.

“We started this ‘green Ethernet' effort with D-Link about four years ago,” explains Vitesse Semiconductor product marketing manager Jason Rock. Vitesse's entry uses its EcoEthernet V2.0 technology, claimed to offer the industry's best power savings for high-density Ethernet switching applications that comply with the IEEE 802.3az EEE standard. These include the VSC 7420-7427 switches and the VSC8512 and VSC8522 SimpliPHY Gbit Ethernet transceivers. The SimpliPHY series employs Vitesse's PerfectReach power savings algorithm that actively adjusts power savings based on the Ethernet cable's length.

“In the Gbit Ethernet mode, our integrated PHYs consume less than 400 mW/ port which is 20% lower than recently announced competitive products. In the Fast Ethernet mode, they consume 50% less power than competing PHYs. Power saving modes such as EEE reduce power consumption by an additional 75% or more,” says Uday Mudoi, Vitesse director of marketing and a participant on the IEEE 802.3az task force.

Broadcom is the one of the first firms to implement the EEE standard with a portfolio of switches that includes Gbit Ethernet switches, ranging from five to eight ports, and a 64-port 10-Gbit Ethernet solution. These chips range from entry level unmanaged switches to enterprise and metro-class switches; single, quad and octal Gbit Ethernet PHYs; dual and quad 10-Gbit Ethernet PHYs; and 1-Gbit Ethernet converged NICs (C-NICs).

Broadcom claims its EEE-compliant products offer energy savings of up to 70% over PHYs without the EEE feature. One such chip Broadcom provides, the BCM54680E PHY, is an eight-port 10/100/1000BASE-T transceiver. All eight transceiver circuits are integrated on a single CMOS chip.

As part of the EEE initiative, Broadcom developed its proprietary AutoGrEEn technology to facilitate the adoption of the standard and provide a faster migration for legacy equipment. This feature lets customers make existing network equipment EEE-compliant by simply changing the PHY devices. Broadcom actually allows users to go beyond the EEE standard with its own Energy Efficient Networking (EEN) initiative.

One of the first to jump on the 802.3az EEE initiative is Hewlett-Packard, with its E5400 zl Ethernet switch line. It consists of six-and 12-slot chassis and associated zl modules and bundles. The foundation for all these switches is a programmable ProVision ASIC that allows for the implementation of demanding networking features such as quality of service (QoS) and security. The products target 10/100 Gbit Ethernet applications. Hewlett-Packard plans to incorporate the EEE feature across multiple devices, including servers, laptops and wireless products.

Realtek is another company among the first to implement the 802.3az EEE initiative with its single-chip RTL8367M-GR 5+2-port 10/100/1000Mbit Ethernet switch controller. The RTL8367M-GR also supports IGMP v1/v2/v3, VLAN translation, double VLAN tagging, ACL, loop detection/prevention, and includes a built-in MCU. It offers advanced QoS and network management functions.

Continue on next page

Teranetics' TN8000 family of 10GBASE-T PHYs is also aimed at 802.3az Ethernet applications. Included are the TN8044 quad-port, TN 8022 dual-port and TN8020 single-port ICs housed in 27×27-mm ball-grid array (BGA) packages. According to Teranetics, the quad-port unit is the smallest of its kind. Each of these chips dissipates less than 4 W/ port for cable distances of 100 m and as little as 2 W/ port for short-reach distances. Recently Teranetics was purchased by PLX Technology.

Aquantia is sampling its AQ1401 10GBASE-T PHY transceiver, which conforms to the 802.3az EEE initiative. Housed in a 27-mm BGA package, the quad PHY chip is designed for radio-frequency interference (RFI) resiliency, and employs a special architecture for low-power (2 W/port for short distances, 3.5 W/port for longer distances).

Finally, Solarflare Communications Inc. is sampling an 802.3az-compliant SFT9104 quad-port 10GBASE-T transceiver. It is said to dissipate just 2.5 W/port at a full 100-m cable length and 1.6 W for a 10-m link. On top of that, the chip's latency time is just 1.5 µsec, about 1µsec less than others on the market.

Looking ahead

Indications are it will be awhile before the full energy savings effects of the IEEE 802.3az EEE standard are felt. Hugh Barass, a technical leader at Cisco Systems, says a complete redesign of the NIC is necessary before the industry can gain maximum benefits from the EEE standard. “We should expect two to three generations before equipment gets the most efficient it can be,” he notes. Many other experts believe it will take five to ten years before this will happen.

Studies have shown that even greater energy savings are possible once high-speed optical Ethernet networks are planned with energy efficiency and savings in mind. However, copper networking is less dear than optical Ethernet networking, so it is likely that efficiency standards for optical Ethernet will come sometime later.

Before focusing on optical standards, industry is more likely to turn its attention to inefficient data transmission redundancy. Many data centers employ redundant links that are always on rather than in a sleep mode. Experts say putting this extra capacity in low-power standby could bring even greater levels of energy efficiency.

The road to the EEE standard

The idea for Energy Efficient Ethernet (EEE) began with a tutorial presented to the IEEE 802 Working Group in 2005. “It was during that presentation that the idea of a project for Ethernet efficiency germinated in a discussion between Ken Christensen and me,” explains Michael Bennett, Chairman of the IEEE P802.3az Task Force. Christensen is professor and director of the undergraduate program in the Dept. of Computer Science and Engineering at the University of South Florida. Bennett is a senior network engineer at Lawrence Berkeley National Laboratory (LBNL).

Numerous discussions over the ensuing months resulted in what's called a Call for Interest (CFI). In November 2006, a panel presentation was given to the IEEE 802.3 Working Group describing the rationale for forming a study group to determine the need for an EEE project, which was followed by a successful CFI vote.

The first meeting of the IEEE Study Group took place in January 2007. The group's aim was to determine whether or not to request a project for EEE. Among the topics discussed were whether there was broad market support for the idea and how to ensure compatibility with existing Ethernet devices.

In September 2007 the P802.3az Task Force formed. The Task Force produced its first draft of the standard in October, 2008 and completed the selection of baseline proposals in March, 2009. After several balloting phases, the IEEE 8021.3az was approved by the Standards Board of the IEEE on September 30, 2010.