PaperlayerView exact revision on arXiv

Source-linked AI summary

A Survey of Green Networking Research

Aruna Prem Bianzino, Claude Chaudet, Dario Rossi, Jean-Louis Rougier

arXiv:1010.3880v1cs.NI

TL;DR

Green networking addresses unnecessary energy consumption in wired networks, where over-provisioning and redundancy conflict with energy objectives. This survey defines the green attribute, identifies enabling paradigms, and classifies the state of the art into four research branches. It finds that Adaptive Link Rate is the most mature branch, while Energy-Aware Applications are the least mature and retain substantial open problems.

  • Problem

    Wired networks are commonly over-provisioned and redundant, creating a need for energy-aware designs that preserve quality of service and reliability.

  • Method

    The paper defines green networking, identifies four enabling paradigms, and surveys and taxonomizes relevant wired-networking research.

  • Results

    The survey categorizes research into Adaptive Link Rate, Interface-Proxying, Energy-Aware Infrastructure, and Energy-Aware Applications, ordered roughly from most to least mature.

  • Takeaways & Limitations

    Energy-Aware Applications and other less explored areas remain important targets for future research and open-problem development.

  • Takeaways & Limitations

    Comparing approaches remains difficult because trace-driven evaluations use different input workloads, motivating agreed standard benchmarks.

Abstract

from arXiv · show

Reduction of unnecessary energy consumption is becoming a major concern in wired networking, because of the potential economical benefits and of its expected environmental impact. These issues, usually referred to as "green networking", relate to embedding energy-awareness in the design, in the devices and in the protocols of networks. In this work, we first formulate a more precise definition of the "green" attribute. We furthermore identify a few paradigms that are the key enablers of energy-aware networking research. We then overview the current state of the art and provide a taxonomy of the relevant work, with a special focus on wired networking. At a high level, we identify four branches of green networking research that stem from different observations on the root causes of energy waste, namely (i) Adaptive Link Rate, (ii) Interface proxying, (iii) Energy-aware infrastructures and (iv) Energy-aware applications. In this work, we do not only explore specific proposals pertaining to each of the above branches, but also offer a perspective for research.

I. INTRODUCTION

Green networking addresses the energy, economic, and environmental costs of wired networking by embedding energy awareness while preserving performance. The survey frames the field's scope, motivations, and opportunities, including infrastructure redesign and service relocation.

  • Data centers and networking infrastructure consume substantial energy because high-performance, highly available equipment requires power-intensive cooling and redundancy.
  • This survey focuses primarily on wired networks because wireless energy studies are sufficiently specific to require a dedicated treatment.
  • Energy savings in wired networks can require trading network performance or redundancy against lower energy consumption.
  • Potential savings include low-power components, renewable energy, architectural relocation, geographically distributed computing, and time-varying service placement.
  • Service migration could follow seasonal or day/night energy patterns, but must preserve service continuity, fault tolerance, and data security.
  • Green networking is defined here engineering-wise as reducing the energy required for a task while maintaining the same performance level.

III. GREEN STRATEGIES

Green strategies counter over-provisioning and redundancy through four paradigms: resource consolidation, virtualization, selective connectedness, and proportional computing. These paradigms reduce underutilized resources or align energy use more closely with utilization, while introducing distributed coordination challenges.

  • Over-provisioning and redundancy increase energy consumption, motivating network designs that add energy awareness without compromising quality of service or reliability.
  • The survey identifies resource consolidation, virtualization, selective connectedness, and proportional computing as four green-networking research directions.
  • Resource consolidation: Resource consolidation adapts active capacity to current demand, for example by shutting down lightly loaded routers and rerouting traffic.
  • Selective connectedness: Selective connectedness lets individual edge resources become idle transparently, whereas consolidation targets shared infrastructure resources.
  • Virtualization: Virtualization places multiple services on one machine to improve hardware utilization and can reduce hardware, energy, cooling, and carbon-footprint costs.
  • Proportional computing: Proportional computing seeks energy consumption that tracks utilization rather than remaining constant, spanning systems, protocols, devices, and components.
  • Proportional computing: Distributed architectures must select and enforce target loads because total energy consumption depends on every device's operating utilization.

IV. TAXONOMY OF GREEN NETWORKING RESEARCH

The survey classifies green networking solutions using decision timing, information scope, network planning role, protocol layer, decision inputs, and evaluation methodology.

  • Classification dimensions: The taxonomy separates online solutions, acting at runtime, from offline solutions, acting before runtime during network design or dimensioning.Online solutions are further divided by whether decisions use local or global information.
  • Classification dimensions: Online strategies are classified as local or global according to whether decisions require information from one node or link versus multiple nodes and links.
  • Classification dimensions: Solutions are also classified by their application to data-link, network, transport, application, or multiple TCP/IP layers.
  • Classification dimensions: Decision inputs may be instantaneous measurements, historical observations, or forecasts combining current and historical information.These input types apply to online solutions, while offline decisions can also use broader planning information.
  • Evaluation methodology: Evaluation approaches include discrete-event simulation, hardware prototyping, and formal analytical or numerical models, each with distinct merits and limits.The survey notes that prototyping can indicate greater research maturity than theory or simulation alone.
  • Scope choices: The classification deliberately favors criteria relevant to the survey rather than network segment or service type, adopting a largely service-agnostic viewpoint.Most reviewed work targets the network edge because it was expected to offer the largest gains.

B. Classification proposal

The survey organizes prior work into branches that target energy use at links, edge interfaces, infrastructures, and applications, while identifying Adaptive Link Rate and Interface Proxying as comparatively established areas.

  • Classification proposal: Table II groups surveyed contributions into branches representing the main energy-aware strategies identified in the literature.
  • Adaptive Link Rate: Adaptive Link Rate reduces energy consumption online in response to low utilization through link-local or network-global techniques.Its strategies follow the proportional computing paradigm and may also use selective connectedness.
  • Interface Proxying: Interface Proxying delegates network-traffic processing from power-hungry mainboard CPUs to low-power NIC components or external proxy devices.The branch targets unnecessary energy waste at edge devices and has been explored by multiple works.
  • Emerging areas: Energy-Aware Infrastructure and Energy-Aware Applications are identified as two further research areas that remain much less explored.
  • Related areas: The survey also briefly covers related areas outside its strict scope, including data centers, computer architectures, and wireless networking.It separately reports measurement and modeling work on energy consumption in networked systems and equipment.

B. Rate switch

Rate-switch research adapts transmission speed to energy conditions, but must balance energy savings against complexity, bursty traffic, quality of service, and implementation constraints.

  • Rate options: Modern technologies offer multiple transmission rates with distinct energy consumption figures, making rate selection an energy-management opportunity.Ethernet spans 10 Mb/s to 10 Gb/s, with higher capacities under development.
  • Energy impact: A 10 Mb/s-to-1 Gb/s NIC rate increase adds about 3 W, while the same switch throughput shift adds about 1.5 W per interface.The NIC increase represents about 5% of overall system energy consumption.
  • Control policies: Rate-control policies have been refined using either the current system state or historical analysis.
  • Optimization: Selecting rates can be formulated as an integer linear program minimizing total energy while forwarding all input traffic, a problem known to be NP-hard.
  • Trade-offs: Sleeping mode has lower management complexity at comparable performance, whereas dynamic rate switching is more robust to bursty traffic and load-estimation errors.
  • Implementation constraints: Rate adaptation faces a quality-of-service trade-off and practical requirements for buffer-based decisions, termination synchronization, and standardized low-energy states.

VI. INTERFACE PROXYING

Interface Proxying preserves end-device connectivity while allowing energy-hungry systems to sleep by delegating filtering and response tasks to NICs or external proxies.

  • Motivation: End devices cannot simply power down because background traffic requires processing, so network functions must be delegated to more energy-efficient entities.Much idle-period traffic can be dropped or handled with minimal computation, including broadcast frames, ARP, ICMP, and DHCP exchanges.
  • Mechanism: Interface Proxying filters unwanted traffic, generates simple responses, or handles more complex application tasks on behalf of sleeping devices.
  • Proxy placement: Proxy functions can run on a NIC’s low-energy processor or on an external device serving one machine or multiple LAN hosts.The approach follows selective connectedness and is categorized as an online local solution.
  • NIC proxying: NIC proxying can power down end systems without disrupting connectivity by waking the full system only for non-trivial packets.According to the survey, it may apply to more than 90% of packets received by an idle PC.
  • Limitations: A reported proxying study omitted analysis of the non-negligible time required for the whole system to wake up.
  • Implementation results: Hardware proxying supports up to 10 Gbps and consumes only 25% of the energy required by the software implementation.The software implementation supports up to 1 Gbps.
  • External proxying: An external switch proxy can answer ARP requests for a sleeping computer until application data traffic arrives.
  • Limitations: Simple proxies can filter broadcast traffic, but properly handling unicast traffic requires significant implementation effort.

VII. ENERGY-AWARE INFRASTRUCTURE

Energy-aware infrastructure research coordinates resources or redesigns architectures to reduce consumption while preserving required network performance and robustness. Approaches range from local collaboration and incremental adaptation to clean-slate optical designs and energy-aware network planning.

  • Infrastructure approaches: Energy-aware infrastructures pursue broader coordination than local device decisions, using shared system-state knowledge for further energy savings.The survey distinguishes individual-device mechanisms from approaches involving wider collaboration.
  • Infrastructure approaches: Two architectural strategies are identified: incremental construction over existing infrastructures and clean-slate redesign of a new architecture.Existing proposals include both approaches, with incremental and clean-slate examples reported in the surveyed literature.
  • Incremental approaches: Incremental infrastructure methods include centralized resource on/off scheduling and global backbone link-rate adaptation through traffic reshaping.Traffic reshaping groups packets destined for the same egress router into bursts, adding end-to-end delay at network ingress.
  • Clean-slate approaches: Clean-slate proposals emphasize optical networking for greater energy efficiency and capacity, while recognizing reduced flexibility because optical networks cannot buffer data.The survey identifies buffering limitations as a motivation for optical burst switching.
  • Energy-aware design: Operational-research approaches add energy consumption to network-design formulations alongside performance and robustness constraints, including tradeoffs with fault tolerance.These formulations evaluate possible working points according to energy consumption, performance, and fault-tolerance characteristics.

B. Energy-aware routing

Energy-aware routing consolidates traffic onto selected links and devices so others can sleep, while preserving connectivity and QoS. The surveyed work formulates this as an offline optimization problem, but important robustness and inter-domain effects remain insufficiently studied.

  • Energy-aware routing: Energy-aware routing aggregates flows over selected devices and links, allowing other network elements to switch off while preserving connectivity and QoS.QoS may be protected by limiting link utilization or maintaining minimum path diversity.
  • Energy-aware routing: Routing weights can achieve flow aggregation, and the resulting problem is a capacitated multi-commodity flow instance handled as an offline dimensioning solution.The survey places energy-aware routing at the network layer and treats it as an offline solution.
  • Coordination and protocols: Energy-aware routing has been discussed in relation to coordinated router sleeping, including implications for fault tolerance and routing-protocol changes.The position paper notes that OSPF treats sleeping links as faulty by default.
  • Optimization methods: ILP formulations and greedy heuristics progressively switch off nodes and links, with selection strategies evaluated in ISP networks containing multi-homed edge nodes.The reported topology is considered a best-case scenario because redundant nodes and links can increase solution efficiency.
  • Open issues: Inter-domain traffic effects and robustness to faults and traffic changes remain unstudied in the surveyed energy-aware-routing literature.These gaps define important boundaries for interpreting current results.

IX. AT THE NETWORK EDGE

Research at the network edge extends energy awareness to computers, data centers, and wireless systems, where operating-system, consolidation, and radio-interface techniques address different energy constraints. Some concepts transfer across environments, but wired and wireless conditions limit direct portability.

  • Scope: Energy efficiency at the network edge concerns end-user computers and data centers, whose architectures differ from network devices and require separate analytical treatment.The survey states that its network-oriented taxonomy is not directly relevant to these systems.
  • Computer architectures: Computer-architecture research includes dynamic voltage scaling, instruction grouping, energy-aware memory allocation, thread migration, and multi-core energy-management policies.These techniques span operating-system and hardware-resource management.
  • Data centers: Data-center research applies selective connectedness and resource consolidation, including work that jointly combines Adaptive Link Rate with network-traffic and server-load consolidation.The survey characterizes data-center efficiency as a more mature and confined field.
  • Wireless networking: Wireless energy awareness is shaped by battery and radio costs, leading mobile devices and sensors toward low-duty-cycle operation.Radio transmission and reception are both described as energy-intensive.
  • Cross-environment transfer: Wireless techniques are not easily transferable to wired infrastructures because wired devices differ in utilization patterns and generally lack the same collision and interference constraints.Some adaptations remain possible, such as Dynamic Voltage Scaling and interface proxying.
  • Cross-environment transfer: Cellular networks can exhibit nighttime over-provisioning analogous to energy-aware infrastructure problems in wired networking.Dense meshes of low-range base stations increase parallel communication capacity, creating excess capacity at night.

X. MEASUREMENT AND MODELS

Measurement and modeling are necessary to identify practical energy-saving opportunities and evaluate green-networking solutions. The survey highlights traffic-pattern measurements, model tradeoffs, software effects, and the continuing lack of representative network-equipment data.

  • Evaluation foundations: Layer interactions may make independently designed energy optimizations counter-productive, motivating coordinated evaluation across link, network, transport, and application layers.The survey presents power modeling and measurements as necessary for assessing green solutions.
  • Traffic and application measurements: Empirical measurements of network and application behavior are required to assess whether green-networking principles are feasible.Traffic patterns determine the opportunities available to proportional-computing and proxying approaches.
  • Traffic and application measurements: Packet inter-arrival analysis motivated sleeping during idle periods, while chatter-traffic and user-presence measurements motivated selective connectedness and proxying.The survey presents measurement as a continuing tool for identifying practical energy-saving opportunities.
  • Energy models: Energy models trade simplicity against accuracy, with alternatives favoring either detailed component-level reporting or a small set of correlated parameters.The comparison concerns measurements of complete systems and components such as CPUs, memory, and hard drives.
  • Software effects: Software choices can alter host energy consumption, as shown by fine-grained measurement of programs using compressed versus uncompressed data.The compressed-data version increases CPU load, illustrating that application design affects energy evaluation.
  • Network-equipment measurements: The field lacks representative network-equipment measurements spanning device types and technologies, limiting comparative energy evaluation.The survey calls for data covering equipment such as DSLAMs, Ethernet switches, and set-top boxes across link technologies.

D. Consumption of networks

Network energy consumption is difficult to measure reliably, and the survey identifies benchmarking and comparable metrics as important needs while organizing green networking research into four branches.

  • Energy-consumption estimates for large networks must account for device power, architectural redundancy, and cooling systems.
  • Existing network-energy models make simplifying assumptions, including fixed cooling overheads, neglected equipment redundancy, and lower-bound estimation.
  • Reliable, representative, up-to-date energy figures remain urgently needed, motivating benchmarking techniques for comparing network solutions.
  • Trace-driven evaluations are difficult to compare because different approaches use different input workloads.
  • Evaluation criteria: Green-networking studies also need consensus on standard performance metrics, since Adaptive Link Rate work reports either total-energy reduction or low-energy-state time.
  • Survey synthesis: The survey defines green networking, presents mainstream paradigms, proposes a taxonomy, and identifies four emerging branches.
  • Survey synthesis: The four branches differ in maturity, with Adaptive Link Rate most explored and Energy-Aware Applications least mature.

A. Adaptive Link Rate

Adaptive Link Rate reduces energy consumption by adapting link rates to traffic, while Interface Proxying extends idle periods through low-power offloading. Energy-Aware Infrastructure addresses network-wide design and management but is less advanced than the former areas.

  • Adaptive Link Rate: Adaptive Link Rate proposals automatically adapt link rates to traffic levels to reduce energy consumption.
  • Adaptive Link Rate: Adaptive Link Rate is well advanced, with an IEEE standardization draft and product deployment expected at the time of the survey.
  • Adaptive Link Rate: Future Adaptive Link Rate research should compare switching strategies at local and global levels while considering perceived QoS effects.
  • Interface Proxying: Interface Proxying uses a low-power intermediate entity to offload network operations and extend idle periods for power-hungry devices.
  • Interface Proxying: Proxying ranges from filtering chatter traffic to handling more complex non-interactive traffic.
  • Interface Proxying: Interface Proxying lacks comparison studies and thorough analysis of wake-up-time effects on QoS; set-top-boxes and subsystem coordination remain open directions.
  • Energy-Aware Infrastructure: Energy-Aware Infrastructure research considers either clean-slate architectures or energy-aware extensions to deployed routing protocols.
  • Energy-Aware Infrastructure: Energy-Aware Infrastructure is less advanced because radically new or gradually evolved architectures are intrinsically complex to deploy.

D. Energy-Aware Applications

Energy-aware applications span software layers from operating systems and network stacks to protocols and application designs, but the field remains broad and underdeveloped. Future progress requires better measurement, programming guidance, performance evaluation, and analysis of interactions among techniques.

  • D. Energy-Aware Applications: Energy-aware software spans low-level operating-system and network-stack changes as well as higher-level instruction and protocol modifications.Examples include tickless kernels, network-stack changes, green Telnet, and green BitTorrent.
  • D. Energy-Aware Applications: The literature covers multiple design dimensions, including kernel versus user level, protocol tweaks versus protocol design, and client-server versus peer-to-peer applications.Despite this coverage, relatively little work has been conducted, and the broad spectrum of Internet applications leaves substantial work remaining.
  • D. Energy-Aware Applications: Energy measurement and modeling introduced a new input to green networking and may support additional, as-yet-undiscovered research branches.The survey presents measurement and modeling as practices with implications beyond the currently identified application approaches.
  • D. Energy-Aware Applications: Progress requires consensus on benchmarking methods, workloads, and metrics, alongside power models that produce reliable results in realistic scenarios.Precise Internet energy modeling remains difficult because of network scale and rapid technological change.
  • D. Energy-Aware Applications: Application-level energy savings must preserve network performance and user experience rather than converting consolidation or selective connectivity into congestion or unreachability.The survey calls for systematic user-perspective evaluation of delay, jitter, security, resilience, and application completion time.
  • D. Energy-Aware Applications: Because green techniques have largely been studied in isolation, their combined impact across applications, devices, links, and routing remains unexplored.The survey highlights possible interactions among energy-aware applications, interface proxying, energy-aware routing, and adaptive link rates.
Loading 1010.3880v1…