An optimization framework for task allocation in the edge/hub/cloud paradigm

TL;DR

This paper tackles optimal task allocation in a streamlined edge/hub/cloud architecture by formulating a complete binary integer linear program. It introduces an extended task flow graph (ETFG) to capture both application structure and system energy/communication models, enabling offline design-time optimization for either latency or energy minimization. The approach is validated on a real UAV inspection scenario and extensive synthetic benchmarks, demonstrating optimal allocations, scalability to large task graphs, and useful design-space exploration across device configurations. The results show the framework can yield minimal latency and energy consumption while respecting device budgets and task precedence, with practical solver times for offline planning.

Abstract

With the advent of the Internet of Things (IoT), novel critical applications have emerged that leverage the edge/hub/cloud paradigm, which diverges from the conventional edge computing perspective. A growing number of such applications require a streamlined architecture for their effective execution, often comprising a single edge device with sensing capabilities, a single hub device (e.g., a laptop or smartphone) for managing and assisting the edge device, and a more computationally capable cloud server. Typical examples include the utilization of an unmanned aerial vehicle (UAV) for critical infrastructure inspection or a wearable biomedical device (e.g., a smartwatch) for remote patient monitoring. Task allocation in this streamlined architecture is particularly challenging, due to the computational, communication, and energy limitations of the devices at the network edge. Consequently, there is a need for a comprehensive framework that can address the specific task allocation problem optimally and efficiently. To this end, we propose a complete, binary integer linear programming (BILP) based formulation for an application-driven design-time approach, capable of providing an optimal task allocation in the targeted edge/hub/cloud environment. The proposed method minimizes the desired objective, either the overall latency or overall energy consumption, while considering several crucial parameters and constraints often overlooked in related literature. We evaluate our framework using a real-world use-case scenario, as well as appropriate synthetic benchmarks. Our extensive experimentation reveals that the proposed approach yields optimal and scalable results, enabling efficient design space exploration for different applications and computational devices.

Figures (7)

• Figure 1: Overview of proposed optimization framework. The task flow graph transformation, including the encapsulated energy model, is described in \ref{['extended', 'subsec:energyModel']}. The formulation of the optimization problem is presented in \ref{['subsec:optimization']}.
• Figure 2: Example of transforming a TFG $G$ into ETFG $G^{\prime}$, considering two different cases of task allocation requirements.
• Figure 3: Real-world use-case scenario: Latency and energy consumption when minimizing overall latency.
• Figure 4: Real-world use-case scenario: Comparison of latency and energy consumption between cases where the optimization objective was the minimization of either latency (O_L) or energy (O_E).
• Figure 5: Overview of random TFG generation and transformation. The generation of random TFGs is presented in \ref{['subsubsec:randomTFGs']}. The transformation of random TFGs into ETFGs is described in \ref{['subsubsec:syntheticParams']}.