Open Research Questions
This page synthesizes insights from our Guided Investigation into three potential directions for sustainable networking research. Each direction connects emerging opportunities in Internet energy efficiency with specific open research questions. Together, they form the foundation for the next stage of this project.
Real-time Carbon-Aware Routing Trade-offs¶
Context¶
Early experiments with traceroute-based carbon mapping suggested that Internet paths could be linked to the carbon intensity of the regions they traverse. Extending this idea to the WAN level raises the question of whether operators could use real-time grid data to favor lower-carbon routes when multiple paths are available.
Open Research Question¶
How effective can intra-domain WAN routing be in reducing carbon emissions when supplied with real-time power and carbon intensity data, and what are the performance trade-offs?
Opportunities and Challenges¶
This approach could deliver carbon reductions without requiring hardware upgrades, simply by integrating existing grid-intensity data into routing decisions. However, as we have seen, practical limitations quickly emerge: current grid data often lacks the geographic resolution needed for hop-level differentiation, meaning distinct routes can appear equally “clean” even when they are not. The problem is compounded by content delivery networks, which localize traffic close to the user, reducing meaningful variation between alternative paths. Beyond data limitations, operators must also contend with performance trade-offs, as changes to routing may introduce latency, jitter, or instability.
Carbon-Aware Traffic Scheduling and Demand Shaping¶
Context¶
A large share of Internet traffic is flexible in timing—examples include software updates, backups, or batch transfers. By shifting such traffic to moments when networks are underutilized or grids are cleaner, overall energy use could be smoothed and carbon emissions reduced.
Open Research Question¶
How can we design and evaluate schemes that defer or reroute lower-priority traffic in response to energy or carbon conditions, and what is the achievable reduction in network energy usage?
Opportunities and Challenges¶
The potential benefits lie in reducing peak provisioning needs while lowering emissions, with parallels to demand-shaping approaches already used in data centers. The main obstacles are practical: accurately classifying traffic priorities, ensuring fairness across users, and fostering adoption at scale so that real gains can be realized.
Network Device Sleeping and Adaptive Topologies¶
Context¶
Routers and links are built for peak demand, but for much of the day many remain underused. With SDN and programmable control planes, networks can be reconfigured on the fly—powering down idle equipment during off-peak or high-carbon periods, and restoring it when demand rises.
Open Research Question¶
What network control strategies (or protocol modifications) are needed to safely power down routers or links during low-demand or high-carbon periods, and how much energy/carbon savings would this yield in practice?
Opportunities and Challenges¶
This line of work promises meaningful savings by adapting active capacity to real-time demand. Yet it must guarantee stability and resilience: links must be restored instantly when traffic surges, routing protocols must handle topology changes seamlessly, and operators must trust that energy savings will not come at the cost of outages.
Next Step¶
These three questions represent complementary strategies for sustainable networking: carbon-aware routing asks where traffic flows, traffic scheduling asks when it flows, and adaptive topologies ask how much of the network must remain active.
For the next phase of this project, our attention turns to network device sleeping and adaptive topologies. This direction was chosen because it combines technical feasibility with significant potential for measurable energy savings, making it a practical entry point for modeling and simulation, as outlined in our Synthetic Data Simulation.