EXECUTIVE SUMMARY
Ethernet has been termed as the most ubiquitous networking interface in the world. Almost all traffic passes through multiple Ethernet links. The point is that majority of the Ethernet links spend time while idle waiting between packets of data transfer, but consuming power at a near constant rate. In fact, on average, 10% of all IT power is consumed by network devices and network interfaces translating to tens of TeraWatts hour per year. A solution lies in the design of green networks which consume minimal power while idle without losing their functionality. In this research paper, Energy Efficient Ethernet is explored as a suitable way of reducing enormous amounts of power consumed by network resources without limiting their functionality.
BACKGROUND
The proliferation of electronics and information technology concepts in day to day living and its apparent ubiquitous nature has increased focus on energy usage and utilization. Programs have been developed that reduce energy use and these have been concentrated highest energy consumers. However, it has come to the attention of the industry that networking infrastructure and devices consume as high as 10% if the total IT consumption. Thus, becoming logically important to explore how networking equipment and infrastructure can reduce energy consumption without affecting their critical functionalities. It is not relevant whether this move is underpinned by concerns over green-house gas emissions that contribute to climate change or energy shortages. This is due to dwindling oil deposit or the requirement to lessen the burden on expensive and delicate energy infrastructures.
In 2006, a taskforce comprising of Cisco and Intel started a movement that became the precursor to the development of energy efficient networks. The movement under IEEE umbrella lead to the formation of Energy Efficient Ethernet project IEEE 802.3az. The project task force considered the options for alternating the Ethernet standard to increase energy savings and allow efficiency. The interfaces targeted by the project included 10BASE-T, 100BASE-TX, 1000BASE-T, and 10GBASE-T. These interfaces are made up of the vast majority of Ethernet installations located at the edge of the networks where opportunities for energy efficiency are limited. The standard also defined Backplane Ethernet interfaces utilized in blade servers and proprietary systems owing to the minimal amount of change required for those interfaces. One of the considerations given due attention is backward compatibility. The new standard should be deployable in networks where the most of the equipment utilizes legacy interfaces and support applications that are already available in the network. One assumption considered in the development was that with the new standard, energy savings might not be realized in older devices as long as existing functions are fully supported. It will facilitate incremental upgrade of existing systems as the proportion of EEE devices increases.
GREEN NETWORKING
It is the practice of designing energy efficient networks and consolidating devices using various technology protocols and standards to reduce power consumption across the networks. Green networking is an offshoot from green concepts observed in every aspect of life. The trend has encompassed IT in general including the data center and the network. EEE represented the beginning of a paradigm shift in networking architecture and information technology in general. Before the inception of this phenomenon, it was accepted that communication devices continue to use energy at the same rate irrespective of the level of their usage. EEE defines the signaling required to save energy during periods the network is idle. Although it does not quantify how much energy is saved, it lays the ground for the development of systems with minimal energy consumption compatible with future developments.
The concept behind this development is that initially, relatively small energy savings will be realized when a comparison is made between idle energy consumption and full rate usage. However, as more systems become compatible with the standard, greater proportions of energy are saved. Early systems will employ simple application of static logic design in the physical layer devices to reduce consumption when data is not present. Physical layer devices (PHYs) take up to 20 to 40 % of the systems power and the static design methods reduce PHY consumption by up to 50%. Therefore, on average, PHY savings are in the span of 20%.
Contrary to initial systems, later generation networking systems will employ aggressive energy saving techniques such as voltage scaling or power islands. These methods are applicable to all of the system silicon extending further the rate if energy savings. With these new aggressive techniques is the requirement for new architectures to support it. The development of an energy saving architecture will propel green concept to a whole new level where as much as 80% of the total worst case energy consumption could be saved.
Devices located at the core of the network will employ a number of techniques to minimize energy consumption. One of the mechanisms is for the devices to enter a deep sleep state while maintaining network link for security purposes only to wake up in response to a network burst. In the deep sleep state, devices may take quit a time to wake up which can be negotiated using the link layer protocol defined in the standard. Edge facing systems are required to support wake time negotiation and extended wake times so that energy savings can be realized. This consideration is put into perspective in the design of future devices so that the energy footprint of the network can be reduced.
The final development is the development of a network architecture characterized by energy efficient control plane solutions. The total energy savings of the systems is the cumulative savings realized from individual EEE devices interconnected in the network. Through the coordination of control policies, the situation is prevented where each of the devices in the network is trying to optimize their individual energy use to create an overall network that is sub-optimal. Though this kind of architecture development cannot be realized in the near future, intense studies and research are underway.However, the underlying factor is the ability of the early systems to exhibit management and negotiation skills that will enable them to participate in a network-wide energy minimization strategy
CORE PRINCIPLES BEHIND EEE
The fundamental principle of energy efficient Ethernet is the communication link that only consume power when packets are being sent. It is in contrast with more wireline communication protocols developed in the last decades which assume a continuous transmission hence, consume a constant amount of power whether they are active or not. The understanding was the link must be maintained at full bandwidth so that data transmission can be undertaken at any time. In the EEE setting, energy savings are recorded where there is a data transmission gap. It is facilitated by a signaling protocol that tells the transmitter of the gap in the data stream so that the link can go idle. The signaling protocol also indicates a resumption of transmission after a predefined delay.
The signal used in the EEE protocol is a modification of the normal idle transmitted between data packets. This signal is referred as the Low Power Idle (LPI) and the transmitter sends it to indicate that the link can go to sleep. After forwarding LPI, the transmitter quit signaling so that the link goes quiescent. Periodically, the transmitter communicates so that the link does not remain for long in the sleep state without refreshing. Finally, when a transmitter wants to resume normalcy, it sends the normal idle signals. After a predetermined period, the link is activated to sent data again. The protocol is defined by two operation times, Ts time to sleep and TW, time to wake.
The signal can be awakened at any time as there is no minimum or maximum sleep interval. This feature allows EEE to operate efficiently and accommodate time of unpredictable traffic. The default time for each PHY is predefined and designed to be equal to the time taken to transmit a maximum length of packets at a given link speed. In the case of 1000BASE-T, the wake time is 16.5 microseconds, almost the same time taken to transmit a 2kb Ethernet frame.
Link management
A refreshed signal is frequently forwarded while the link is idle for a number of reasons. The first is the traditional link pulse witnessed in traditional Ethernet. The core function of the refreshed signal is to notify both partners that the link is present and allow for immediate notification in case a disconnection is reported. The frequency of refreshing that is usually greater than 100Hz, prevent a scenario where one link partner is disconnected and another joining without a link failure. This feature maintains compatibility with security requirements that depend on continuous connectivity and prompt notification whenever a link is broken.
Another reason for maintaining refresh signals is to enable high layer applications to understand that the link is present to preserve network stability. Al alteration of the power levels should not cause a disconnection, link flip, network reconfiguration or client changes. Finally, a refreshed signal serves the purpose of testing the channel and allowing the receiver to adjust to the channel characteristics. For high-speed links, this feature is essential to allow rapid transition to high-speed data transfer without impacting on the integrity. It has been noted that refresh signal is designed for each PHY type to facilitate adaptation to supported medium.
DEEP SLEEP NEGOTIATION
EEE standard provides the ability to negotiate back to wake time. As earlier noted, a specific wake time is defined for each PHY type and is tailored to be the same as the time taken to transmit a maximum length packet for the particular PHY utilized. In some instances, the devices may be designed to go into longer sleep time, requiring more time to transition to full functionality. Examples such devices are PCs or server network interface cards that render the whole network into sleep and only be awakened by network activity. Another example is systems with large memories that eliminate power from their memory systems in order to minimize power usage.
Thus, a requirement for deep sleep is negotiation. A receiver may exhibit more than one level of sleep that requires different wake times. In the same way, the transmitter may exhibit a limit to the depth of buffering that it can support in order to guarantee holding the packets it arrives until the receiver is ready. A negotiation is required for these and takes place via a link-layer discovery protocol defined by the IEE 802.1AB. Since the standard is supported by most networking equipments, there is no major burden placed on the system to include wake up time negotiation. The negotiation is in tandem with the required sleep mode of the receiver with the optimum delay that the transmitter can hold for maximum energy savings for the required performance. A negotiation can be established for each direction so that asymmetric wake up time is achieved. Also, negotiation is a factor of the policies or conditions of control. For this case, in most cases, sophisticated networks support renegotiation based on control policies derived from network management systems as well as balancing the use of shared buffers between multiple ports so that edge devices can derive the most energy saving. An example of such a system is the Cisco EnergyWise. An example of negotiation is presented below.
EEE deployment
Every PHY that is EEE-compliant advertises its capability via auto negotiation when it is connected to a link. If the link partner is not EEE compliant, the link operates in legacy mode and is similar to two peers before EEE deployment. It implies that EEE-capable systems do not compromise existing networks. They only derive benefits as more and more of them are added into the network as devices or endpoint. This way, more energy savings are recorded. The caveat to this is that early systems may only support the default wake up time operation and the less complex EEE policies. It may imply a limitation on energy savings during the initial phase of deployment. However, as the systems are upgraded and more advanced systems are deployed, wake up time negotiation becomes applicable to facilitate deep sleep. Also, multiple EEE policies come into effect to match sleep and wake behaviors of applicable systems. The result is more efficiency and considerable energy savings.
Use of EEE to deliver services
Energy Efficient Ethernet is defined to allow higher layers services to run over EEE links without degrading networks. One assumption applied in the definition of EEE is that the link can enter power saving mode but still maintain link integrity and guarantee full functionality. The higher layer functions are not dependent on the changes occurring between data and LPI, rather, higher layer functions use management information available from the EEE function to control negotiation time. Since most EEE interfaces default to a wake up time equivalent to the delay of maximum packet length at the target link speed, the impact is minimal on applications as similar delays is witnessed in Ethernet packet through store and forward switching operations. Applications that necessitate maximum latency in LAN configurations such as IP telephony can function without limitations over many EEE links utilizing the same default wake times. For such applications, network administrators should pay close attention to deep sleep operations in important part of the network. For applications designed to operate outside the LAN and in the internet or WAN, they are tolerant of very high latency. As a result, deep sleep negotiation will not be an issue.
Finally, some applications are over sensitive to latency. For instance high-performance computing may respond to latency experienced by inter-processor communication or synchronization of traffic. The same goes with financial trading applications that require latency, much lower than maximum packet delay, and hence the requirement for cut-through switching. A custom operational scenario provided by energy control applications such as Cisco EnergyWise is adopted where EEE functions are disabled when such sensitive applications are running and activated when they are dormant to realize energy savings. In the case of a data center, network-wide control applications that discern some performance degradation on large file transfer as per the particular EEE profile is desirable. The applications function on a peak and off-peak hour schedule to mitigate the effect.
EEE at the edge
The most justification of EEE is realized from typical use of edge devices. Even during peak operations, most client computers utilize network connections with infrequent bursts. It is where EEE is most applicable. During off peak modes, client devices are in sleep or hibernate modes and the network interface is inactive but more often expected to be quickly awakened by a demand from a remote request. This requirement limit that the client must be connected to network serviced constraint the ability of the device to go into low power modes. To solve the situation, a delegation mechanism of the connectivity maintenance to a subsystem is used. The delegation is referred as a network proxy and is being developed. The application of network proxy and EEE to edge devices ensures that the client remains connected to network resources without keeping the network interface at full power.
The edge device can also utilize wake time negotiation when it transits to the low power state. This is such that when packets are sent to the client from a remote device such as wake-on-LAN, the negotiation permits the client sufficient time to transit out of the low power state to facilitate processing of the incoming packet. It increases reliability and performance of service that depend on remote wake up functionality. In traditional Ethernet links, speed downshifting is used when a device such as PC enters a sleep state. For example, 1Gbps to 10/100Mbps downshifting is achieved for 1000BASE-T. With EEE, PCs can save power similar to that in speed downshifting but with a much faster transition period to the active state. The faster link wake up time provides enhanced user experience and faster transition from sleep to an active state.
EEE is finding applications in some of the industry devices. For instance, Cisco Catalyst 4500E Switches is using EEE. A test process simulating normal operation of desktop/laptop was conducted using Cisco Catalyst 4500E system with 384 1000Base-T ports in a lab. A significant 141 watt power reduction was recorded for the 384-port system. With 191 EEE-established links, a power saving of 0.74 watts was recorded. It is a significant portion of power that could extend to terawatts if computed cumulatively in a normal organizational setting.
CONCLUSION
The application of EEE phenomenon for aggressive power savings in networks is feasible. The phenomenon will contribute to green networks working at the same or enhanced operational speeds and efficiency. When the concept is applied at the whole system, the range of power savings is enormous. The development of greener networks takes place in phases. EEE represents the initial phase and as improvements are made in devices, protocols and architectures, more energy savings are realized. Though it might take a while, the benefits outweigh the steps taken.
References
Addis, B. C. (2014). Energy management through optimized routing and device powering for greener communication networks. . IEEE/ACM Transactions on Networking (TON) , 22(1), 313-325.
Bolla, R. B. (2011). Energy efficiency in the future internet: a survey of existing approaches and trends in energy-aware fixed network infrastructures. . Communications Surveys & Tutorials IEEE , 13(2), 223-244.
Cisco, I. (2010). IEEE 802.3az. Energy Efficient Ethernet. Building Green Networks. IEEE .
Nordman, B. &. (2009). Greener PCs for the Enterprise. . IT professional , 11(4), 28-37.
Yano, Y. (2012 February). Take the expressway to go greener. . In Solid-State Circuits Conference Digest of Technical Papers (ISSCC), (pp. pp. 24-30). IEEE.