MHeedra: Putting Duplication-Enabled Task Scheduling Within Heterogeneous Multi-User Edge-Cloud Platforms to Work

Meeting task performance requirements within edge-cloud computing platforms is difficult due to heterogeneous processing, transmission capabilities, and the multiplicity of optimization opportunities. Edge-cloud platforms fill the computing continuum gap and alleviate data-locality performance issues from offloading tasks. Indeed, additional computing resources disseminated across the network but close to the data source help decouple the inherent dependency of cloud computing on best-effort transmission links. Efficient use of such platforms requires complex placement techniques leveraging task duplication to reduce traffic while accelerating response times. In addition, to be fully operational, the task placement strategy should intrinsically support multi-user and tiered platforms. In this work, we designed MHeedra, a modified{Mixed-Integer Linear Programming (MILP)} model-based framework, further adapted into an Ant Colony Optimization (ACO) metaheuristic, capable of scheduling offloading requests, issued by numerous clients, to heterogeneous interconnected edge-cloud computing resources. modified{The scheduler processes these user-issued requests and assigns the encapsulated tasks to one or more edge-cloud devices or nodes, supporting task duplication to minimize delay based on the expected completion time QoS metric. For practical implementation, we designed MHeedra-ACO an $epsilon$-greedy ACO algorithm derived from our optimal model. To evaluate our framework, we implemented a parameterized instance generator to simulate real edge-cloud platforms and workloads and compare our metaheuristic to CPLEX optimal results.} Our experiments show that MHeedra-ACO matches modified{$63.33%$} of the optimal solutions computed by CPLEX, considering all runs across all instances. Moreover, the use of duplication leads to a delay improvement of up to modified{$1.45times$} compared to non-duplication scheduling.