EDD-NSTE: Edge Data Distribution as a Network Steiner Tree Estimation in Edge Computing

TL;DR

The paper tackles the problem of distributing application data to edge servers under a vendor-defined latency constraint $d_{limit}$ while minimizing total transmission costs, combining high-cost cloud-to-edge and lower-cost edge-to-edge transmissions through a Steiner-tree perspective. It first refines an IP formulation for Edge Data Distribution (EDD) and then introduces EDD-NSTE, an $O(k)$-approximation algorithm based on network Steiner tree estimation, leveraging metric closure, Triple Loss Contracting, and rooted Steiner-tree refinement under $D_{limit}$ (with $K=D_{limit}$). Theoretical guarantees bound the final cost to a constant factor of the optimum, specifically $Cost(G_{final}) \\le \\frac{11}{6}\left(\frac{2\gamma}{K} + 1\right) Cost(G_{opt})$, while empirical evaluation on the EUA dataset shows EDD-NSTE outperforms Greedy Connectivity, Random, and the state-of-the-art EDD-A by substantial margins (up to 86.67%). The work delivers a scalable, practical strategy for data placement in edge networks, enabling latency-sensitive applications to reduce cloud-to-edge and inter-edge costs without violating latency budgets. Key ideas include using the metric-closure Steiner framework, rooting the Steiner tree at the cloud, and pruning to meet $D_{limit}$ while maintaining cost efficiency.

Abstract

Edge computing is a distributed computing paradigm that brings computation and data storage closer to the user's geographical location to improve response times and save bandwidth. It also helps to power a variety of applications requiring low latency. These application data hosted on the cloud needs to be transferred to the respective edge servers in a specific area to help provide low-latency app functionalities to the users of that area. Meanwhile, these arbitrary heavy data transactions from the cloud to the edge servers result in high cost and time penalties. Thus, we need an application data distribution strategy that minimizes these penalties within the app vendors' specific latency constraint. In this work, we provide a refined formulation of an optimal approach to solve this Edge Data Distribution (EDD) problem using Integer Programming (IP) technique. Due to the time complexity limitation of the IP approach, we suggest an O(k) approximation algorithm based on network Steiner tree estimation (EDD-NSTE) for estimating solutions to dense, large-scale EDD problems. Integer Programming and EDD-NSTE are evaluated on a standard real-world EUA data set and the result demonstrates that EDD-NSTE significantly outperforms with a performance margin of 86.67% over the other three representative approaches and the state-of-the-art approach.

Figures (15)

• Figure 1: An industry example
• Figure 2: EDD scenario with 10 edge servers
• Figure 3: EDD example to demonstrate dlimit
• Figure 4: Optimal solution using integer programming
• Figure 5: Steiner Tree estimated by Algorithm 1