Energy-aware Distributed Microservice Request Placement at the Edge

TL;DR

Energy-aware decentralized request placement for edge-based microservices is formulated as a Traveling Purchaser Problem and solved via an ILP using two energy objectives, $E^O$ and $E^M$. The work shows that the choice of energy metric materially affects placement decisions and energy outcomes, especially under load heterogeneity, and demonstrates the trade-off between energy savings and completion time within deadline constraints. The evaluation reveals that $E^O$ can exploit farther, energy-favorable devices at the cost of longer completion times, while $E^M$ tends to favor load consolidation and closer devices, with differences amplifying as system heterogeneity grows. However, ILP-based optimization incurs tens to hundreds of milliseconds, indicating a need for fast heuristics or hybrid strategies for real-time edge deployments, while providing a strong optimality reference for future approaches.

Abstract

Microservice is a way of splitting the logic of an application into small blocks that can be run on different computing units and used by other applications. It has been successful for cloud applications and is now increasingly used for edge applications. This new architecture brings many benefits but it makes deciding where a given service request should be executed (i.e. its placement) more complex as every small block needed for the request has to be placed. In this paper, we investigate decentralized request placement (DRP) for services using the microservice architecture. We consider the DRP problem as an instance of a traveling purchaser problem and propose an integer linear programming formulation. This formulation aims at minimizing energy consumption while respecting latency requirements. We consider two different energy consumption metrics, namely overall or marginal energy, to study how optimizing towards these impacts the request placement decision. Our simulations show that the request placement decision can indeed be influenced by the energy metric chosen, leading to different energy reduction strategies.

Figures (6)

• Figure 1: The considered edge system.
• Figure 2: The edge service considered for evaluation
• Figure 3: Average overall energy for a given request with different edge device loads.
• Figure 4: Average overall energy for a given request with varying function instance availability and infrastructure load levels.
• Figure 5: Average completion time for a given request with varying function instance availability and infrastructure load levels.