Scope
The Internet Engineering Task Force (IETF) is the leading standardisation body concerning the Internet and Internet-connected systems. The increased attention to green networking has led to two Internet Draft documents exploring this topic in depth: draft-irtf-nmrg-green-ps-02 and the current document. These documents contain ideas and considerations rather than proposing a normative standard. This document states explicitly that it is intended to spark discussion about what metrics will be useful to reduce the carbon footprint of networks.
In particular, it explains the need for network instrumentation to assess the “greenness of networks”. It proposes several sustainability-related metrics associated with a network, its equipment, and the services provided over it. These metrics can form the basis for optimising energy efficiency, carbon footprint, and environmental sustainability of networks and services. This discussion is a foundation for later IETF standardisation activities, such as defining related YANG modules and protocol extensions. Metrics in the current document are independent of representation as part of an (e.g., YANG) data model.
Summary
The proposed metrics are categorised according to the subject to which they apply:
- Device/equipment level: the device as a whole or its components (e.g., line cards or individual ports).
- Flow level: as per [RFC7011], a flow is “a set of packets or frames passing an Observation Point in the network during a certain time interval”. The set is defined by some common properties of the packets (e.g., destination IP address, next-hop IP address).
- Path level: sequence of links between source and destination.
- Network domain level: aggregation to network domain as a whole (i.e., system level).
A distinction is made between primary metrics (directly measured) and derived metrics (computed from multiple factors). Conversion metrics are based on conversion from other metrics by some factor (e.g., computing CO2e impact from energy, using the energy mix to compute the factor).
The text below enumerates the metrics proposed in the document, categorised by subject, as mentioned above. The metrics for each subject category are grouped into different sets, and individual metrics per set are included when applicable.
Device equipment level metrics
Energy ratings: provided by the manufacturer (e.g., data sheet) or measured beforehand (e.g., in the test lab or when deploying). It may apply to the device as a whole and/or to subcomponents. Individual proposed metrics:
- Power drawn when idle
- Power drawn when fully loaded
- Power drawn at various loads (utilisation percentage)
Primary metrics: actual power during operation. Individual proposed metrics:
- Current power drawn
- Power consumed since system start (or module insertion or port activation)
- Power consumed for the last minute
Derived metrics: primary metrics per unit of traffic. Individual proposed metrics include:
- Current power consumption / kB (or GB)
- Current power consumption/packet
Note that these might not reflect actual utility (e.g., power consumption per minute of video would.)
Incremental energy usage: the (expected) energy usage when adding load (i.e., idle power usage is not amortised). Individual proposed metrics include:
- Incremental power consumption per MTU-sized packet.
- Incremental power consumption per GB
Green metrics beyond energy consumption: include the “sustainability rating” of the power source or mix and the “sustainability tax,” e.g., to account for cooling. Both can be combined into a “pollution factor,” and both will fluctuate over time and might be approximated. Individual proposed metrics include:
- Power Sustainability Factor (PSF), also known as Carbon Intensity or Electricity emission factor.
- Deployment Sustainability Factor (DSF) to attribute share of deployment’s overall power consumption to individual network devices.
Which can be used to derive “sustainability weighted” metrics.
- Current Sustainability-Weighted Power Draw
- Current Sustainability-Weighted Power Draw / GB
- Incremental Sustainability-Weighted Power Draw / MTU-sized packet
Which, in turn, can optionally be converted to approximate CO2 emissions.
Virtual energy metrics: attributing actual power consumed to virtualised entities (e.g., soft switches, VNFs, or networking slices) is a non-trivial task and out-of-scope for the current document. VMware has proposed metrics to attribute energy to VMs.
Flow level metrics
Energy consumption related to a flow at a given device: considering the flow going through a single device. Individual proposed metrics:
- Amortised energy consumed over the duration of the flow: the share of the flow’s energy consumption over the total energy consumption of the device over the duration of the flow.
- Incremental energy consumed over the duration of the flow: the difference between the amount of energy consumed with the flow and the amount of energy that would have been consumed without the flow (hard to determine in practice).
Energy consumption related to the entire flow path: aggregate the flow’s impact on carbon footprint over the entire flow path. This is difficult to assess with reasonable accuracy due to, e.g., load balancing, packet replication, elimination and ordering, and packet loss. Furthermore, tracing actual routes may be difficult. These metrics can be weighted with PSF/DSF as well.
Path level metrics
Energy/power/sustainability related to a path: could be used as cost factor in routing decisions to enable carbon-aware routing. Individual proposed metrics:
- Energy rating of a path: computed as a function of energy ratings or Power Sustainability Factors (PSFs) of the different hops along the path. For example, for PSFs, the function could be max, sum or average.
- Current power consumption across a path, also known as the Path Energy Traffic Ratio, could be computed as the sum of the current power per packet (kB, etc.) for each of the hops along the path.
- Incremental power consumption for a packet over a path: as above, but for incremental power.
Similar to flow path-related metrics, obtaining these metrics may be challenging.
Network domain level metrics
Energy related to the network as a whole: requires normalising service and traffic volume. Individual proposed metrics:
- Total energy consumption (MWh): entire energy consumption attributed to the network.
- Electricity from renewable sources (%): percentage of total energy consumption that comes from renewables.
- Network energy efficiency (MWh/PB): relating total energy consumption to utility derived from the network, as measured by the total amount of data being transmitted.
- Energy efficiency rating (EER): the ratio between the net energy consumption of networking devices and the total energy consumption.
Observe that the term “power” in this document is often used when “energy” (i.e., accumulated power over a time period) is intended.
After discussing the metrics, the document closes with a number of considerations and controversies:
- From a user perspective, attributing energy usage to individual usage may create awareness, but how do we do this accurately/fairly and how do we use this information (e.g., energy-aware charging may be rejected like volume-based charging)?
- From a holistic perspective, aspects outside the network should be considered, like the abovementioned “sustainability tax”.
- The embodied carbon footprint of networking equipment and end-systems (end-user terminals, cloud servers) should be amortised over their lifetime. Also, the energy consumed in performing replacement and disposal could have an impact, as could the energy consumed when making software (hard to measure).
- Imprecision and uncertainty could lead to using intervals rather than single values as metrics.
- Metrics used to optimise must be correct; therefore, certification by a trusted authority may be needed.
- Explicit reference is made to ETSI TC EE (ES 203 228, 202 706-1, 202 706-2, 303 215 , 203 184, and 203 136).
- A basket of metrics may be needed to create sufficient nuance, metrics complementing each other to sketch a more holistic picture. Comparing deployments may be like comparing apples and onions, but benchmarking might help (out-of-scope).
Relevance for EXIGENCE
This document is mostly about defining metrics and therefore directly relevant to energy metrics. Additionally, the topic attribution is very briefly touched upon, which might bear relevance to energy measurements. More importantly, this is ongoing work and therefore provides an opportunity to contribute to standardisation.
