Workload Distribution with Rateless Encoding: A Low-Latency Computation Offloading Method within Edge Networks

TL;DR

This work tackles low-latency, reliable computation offloading in mobile edge networks with heterogeneous, time-varying nodes and stragglers. It introduces REDC, a framework that couples rateless encoding with adaptive workload distribution, guided by M/G/1 queuing and EWMA-based state feedback, and employs a Quick Launch Strategy to start computation early. Key contributions include integrating encoding and scheduling into a unified, adaptive offloading policy, modeling decoding and overheads via $\epsilon_{\mathrm{dec}}$ and $\epsilon_{\mathcal{S}}$ with $\Gamma = 1+\epsilon_{\mathrm{dec}}+\epsilon_{\mathcal{S}}$, and demonstrating performance gains under heterogeneous and unstable edge conditions. The results show reduced execution time and improved resilience to stragglers, suggesting significant practical impact for edge-cloud workflows requiring fast, reliable matrix-multiplication workloads.

Abstract

This paper introduces REDC, a comprehensive strategy for offloading computational tasks within mobile Edge Networks (EN) to Distributed Computing (DC) after Rateless Encoding (RE). Despite the efficiency, reliability, and scalability advantages of distributed computing in ENs, straggler-induced latencies and failures pose significant challenges. Coded distributed computing has gained attention for its efficient redundancy computing, alleviating the impact of stragglers. Yet, current research predominantly focuses on tolerating a predefined number of stragglers with minimal encoding redundancy. Furthermore, nodes within edge networks are characterized by their inherent heterogeneity in computation, communication, and storage capacities, and unpredictable straggler effects and failures. To our knowledge, existing encoding offloading approaches lack a systematic design and unified consideration of these characteristics. REDC addresses these issues by adaptively encoding tasks, then distributing the workload based on node variations. In the face of unpredictability failures, the rateless encoding adaptation provides resilience to dynamic straggler effects. Considering the node heterogeneity and system status, tasks are offloaded to optimal subset "valid" nodes. Load distribution decisions are made based on updates to queuing theory modeling through state feedback. The REDC framework is applicable to EN by improving resource utilization and reducing task sequence execution delays. Experimental results demonstrate our method's effectiveness and resilient performance, maintaining efficacy even in the presence of unstable nodes.

Figures (11)

• Figure 1: Example scenario for computationally intensive workloads offloading Demand. In our context, User elements continuously generate computationally intensive tasks, which are offloaded to the edge computing nodes (Local DN) and Cloud server for distributed processing. This type of computational offloading is essential to optimize the utilization of resources and improve the overall system performance.
• Figure 2: Illustration of coded distributed computing: 2D-MDS Code ($(3,2)^2$ Product Code)lee2017high in an example with $N = 9$ workers that can each store half of $\mathbf{A}$ and half of $\mathbf{B}$, where $\text{a} = [\text{a}1,\text{a}2,\text{a}3] ^{\text{T}}= [\mathbf{A}_1,\mathbf{A}_2,\mathbf{A}_1+\mathbf{A}_2] ^{\text{T}}$,$\text{b} = [\text{b}1,\text{b}2,\text{b}3]= [\mathbf{B}_1,\mathbf{B}_2,\mathbf{B}_1+\mathbf{B}_2]$.
• Figure 3: Overview of the Rateless Encoding Distributed Computing (matrix multiplication) framework. a. The system includes controllers, encoders, schedulers, workers, and decoders; b. Focus on matrix operations and show the operation process; c. Consider calculation execution time including encoding time, data upload time to computing nodes, and operation time. Calculation result return time, decoding time; d. The computing tasks we consider are executed sequentially, and the computing tasks of the workload arrive randomly.
• Figure 4: bipartite graph.
• Figure 5: Job execution time.