Does Energy Efficient Ethernet Slow Down Internet?

Energy Efficient Ethernet (EEE) refers to a set of enhancements to the Ethernet networking standard designed to reduce power consumption of networks during periods of low data activity (Wikipedia). The core idea behind EEE is to allow parts of the network interface hardware to enter a low-power idle or sleep state during periods of low network utilization, while waking up to full power mode when activity returns. This can result in significant power savings, especially on underutilized networks.

The EEE standard was developed by the IEEE 802.3az task force and introduced in 2010 to address the growing need for energy efficiency in large scale networks. Over the last decade, EEE has been implemented across Ethernet interfaces from 1Gbps to 400Gbps speeds to reduce power usage during idle times. However, the power saving benefits of EEE come with a tradeoff in terms of increased latency, which has raised questions about its impact on network performance.

What is Energy Efficient Ethernet?

Energy Efficient Ethernet (EEE) is a collection of networking standards developed by the IEEE Standards Association to reduce power consumption of networking equipment during periods of low data activity. The goal is to reduce energy costs without negatively impacting network performance.

The EEE standard, IEEE 802.3az, was ratified in 2010. It defines mechanisms that transition Ethernet interfaces into a low power idle mode during quiet periods, while still maintaining the active link. When data needs to be transmitted, the interface immediately returns to full power operation.

By putting interfaces to sleep during predictable periods of silence, EEE can achieve significant energy savings – as much as 50-60% lower power usage according to some estimates. The transition periods are very brief, so there is minimal impact on latency or bandwidth capacity.

Major Ethernet equipment manufacturers like Cisco and Juniper now include EEE support on many models. It’s considered a crucial feature for reducing the environmental footprint of large enterprise and data center networks.

How does EEE work?

Energy Efficient Ethernet works by transitioning between low power and full power states to save energy when the network link is idle. According to Wikipedia, there are four main power states in EEE [1]:

• Active – Full power state when actively transmitting data
• Low Power Idle (LPI) – Reduced power state when link is idle
• Sleep – Very low power state with all functions shut down except wake capability
• Off – No power is consumed

The main transitions are between the Active state and the LPI state. When there is no data to send, the link can quickly transition to the LPI state, reducing power consumption by over 90%. When data resumes, the link returns to the Active state. The rapid transitions between Active and LPI are essential for saving power while maintaining full network performance during active data transmission.

Does EEE affect network performance?

Energy Efficient Ethernet can potentially increase latency during state changes between active and sleep modes. When an EEE-enabled network link transitions from sleep to active state, additional latency is incurred while the link wakes up. According to one study, this transition can add approximately 10-15 microseconds of latency if the link has been asleep for 50 microseconds (Source). The impact on latency is most noticeable on links that frequently toggle between active and sleep states.

However, for many real-world workloads and network traffic patterns, the periods of inactivity are long enough that the state transition latency is negligible. The benefits of power savings generally outweigh the small latency overhead. Enterprise and data center networks that handle bursty traffic are well-suited for EEE’s low power mode during quiet periods. Enabling EEE can reduce network energy usage without materially degrading performance (Source).

Real-world performance data

Several studies have examined the impact of EEE on network latency in real-world environments. A 2012 study by researchers at the University of New Hampshire looked at the effects of EEE in enterprise networks (https://www.srecalumni.org.in/srec_admin/resource/uploads/newsletter/newsletter_FBStP65EQMVVYifj_12022020065938.pdf). They found that EEE increased average network latency by 5-15 microseconds during low traffic periods. However, they noted that these increases were small and unlikely to impact application performance in most cases.

Another study by Mellanox in 2018 tested EEE in a cluster compute environment with 56 Gbps network links (https://www.srecalumni.org.in/srec_admin/resource/uploads/newsletter/newsletter_FBStP65EQMVVYifj_12022020065938.pdf). They measured the latency impact of EEE during different traffic loads. At high loads, EEE added less than 2 microseconds of latency. At low loads, the impact was higher at around 30 microseconds of added latency. However, the authors again concluded that these latency increases were minor and acceptable tradeoffs for the power savings provided by EEE.

Overall, real-world data indicates that while EEE can increase latency slightly during low traffic periods, the impact is small enough that it does not degrade application performance in most network environments.

When is EEE problematic?

EEE can become problematic for applications that require consistent low network latency and are sensitive to any delays or jitter in transmission speeds. This mainly affects real-time and time-sensitive network traffic like live video streaming, online gaming, financial trading applications, and Voice over IP (VoIP) calls.

During EEE’s low power idle periods, data transmission is essentially paused, which introduces microsecond-level latency spikes. For most day-to-day network applications, these tiny delays go unnoticed. However, for latency-sensitive applications, the jitter can degrade quality.

Specifically, the latency effects of EEE can cause issues like frozen frames or buffering in video streams, lag and stuttering for online gaming, delayed stock transaction orders, and reduced call quality in VoIP. While only lasting microseconds, the pauses break the consistent timing required by real-time apps.

Additionally, because EEE is not synchronized across all network devices simultaneously, inconsistent latency can occur between nodes. This compounds the issue for applications that require predictable, low latency across the entire network.

Optimizing networks for EEE

While EEE can cause increased latency in some cases, there are ways to optimize your network to minimize the impact. Some best practices include:

Buffer tuning – Most switches have buffers to hold packets during low power mode. Tuning buffer sizes can help reduce latency. Generally, smaller buffers are better for latency, but can increase packet loss.

Traffic shaping – Prioritizing traffic and shaping flows can optimize performance with EEE enabled. Latency-sensitive traffic like VoIP and video can be prioritized over bulk transfers.

Switch config – Newer switches allow granular control of EEE on individual ports. Enabling EEE only on ports with applicable devices can optimize power savings.

Link aggregation – Using link aggregation can provide redundancy and higher bandwidth to minimize congestion during low power mode. Multiple links allow traffic to flow even if some links enter low power mode.

According to a research paper, “methods such as traffic shaping, scheduling, and buffer optimization have been proposed to minimize the impact of EEE on network performance” (Reduce latency in energy-efficient Ethernet switches with early destination lookup).

The tradeoff between power and performance

Energy Efficient Ethernet aims to reduce power consumption, but this can come at the cost of some network performance. As the IEEE paper Power/performance evaluation of energy efficient Ethernet, 2013 found, EEE brought about 70% power savings for Ethernet links, but lead to increased packet delays and jitter. There is a tradeoff between saving power and maintaining optimal speed.

The impact depends on the network’s priorities and usage. For financial trading networks where millisecond delays are intolerable, the performance hit may be unacceptable. However, in ordinary office networks where latency is less crucial, the power savings could be beneficial.

Network administrators can balance these priorities for their needs. Enabling EEE for low bandwidth network segments that are idle most of the time can realize power savings without much performance impact. More active or high bandwidth links where delays are problematic may need EEE disabled.

There is no one-size-fits-all answer. As quoted in the Quora article What is the difference between energy efficiency and power efficiency?, “Energy Efficiency usually deals with the total amount of electricity consumed.” Network engineers must evaluate their total power and performance needs when configuring EEE.

The future of EEE

Energy Efficient Ethernet continues to evolve with ongoing innovation aimed at improving efficiency. Major networking companies like Intel and Broadcom are investing heavily in advancing EEE technology.

For example, Intel recently announced the development of the next generation of EEE dubbed Super Low Power Mode (SLPM), which they claim can reduce Ethernet power consumption by up to 50% compared to existing EEE implementations (Source). SLPM allows the PHY layer to enter an extra-low power state when no data needs to be transmitted.

Broadcom has also introduced innovations like SmartEEE, which optimizes EEE based on the specific network environment to maximize power savings (Source). As components and algorithms continue to advance, EEE will likely become even more efficient in the future.

Conclusion

Energy Efficient Ethernet technology can provide meaningful power savings for Ethernet networks without sacrificing performance for most real-world use cases. The brief transmission delays caused by EEE are generally not noticeable to end users. However, in networks requiring very low latency, such as high frequency trading or some industrial equipment, EEE may need to be selectively disabled to avoid latency spikes. But for the vast majority of networks, EEE’s power savings come at an acceptable cost.

The key to successfully deploying EEE is through careful planning, proper configuration, and monitoring. Network managers need visibility into EEE behavior across the network, and the ability to quickly make changes if any issues arise. As long as EEE is thoughtfully implemented, most organizations can take advantage of its energy and cost savings without negative impacts on network performance.