This work considers the problem of cost-effectively mapping a swarm of soft real-time stream processing applications with moldable-parallel tasks to multicore resources in the device-edge-cloud continuum, consisting of mobile devices, edge resources and cloud resources, and presents several offline algorithmic solutions.
Abstract
We consider the problem of cost-effectively mapping a swarm of soft real-time stream processing applications with moldable-parallel tasks to multicore resources in the device-edge-cloud continuum, consisting of mobile devices, edge resources and cloud resources. We leverage flexibility from different parallelization degrees and frequency levels (DVFS) for the tasks, keeping application throughput constraints and communication bandwidth limitations while minimizing overall cost (including device/edge resource energy, communication cost and cloud resource renting). We present several offline algorithmic solutions with a global view of the environment: an integer linear program (ILP) extending the crown scheduling approach for multi-layer distributed systems, a variant leveraging symmetries in application and system structure, and a greedy heuristic algorithm. We also expanded the problem formulation to consider the dynamic joining of application task graphs, introducing a dynamic approach based on the proposed ILP and greedy heuristic algorithm. Our experimental evaluation for several real-world and synthetic scenarios shows that the time required for solving the scheduling problem to cost-optimality by the ILP is feasible for nontrivial scenarios. The heuristic achieves about 3% worse cost efficiency on average, yet operates much faster (by 1–2 orders of magnitude), allowing to scale up the problem size more than the ILP approach. The symmetry-folding applied to the ILP approach improves its optimization time by about 1 order of magnitude, at the expense of less than a 5% increase in cost compared to a non-folded static ILP solution. The heuristic is likewise accelerated by leveraging symmetry, though to a minor extent. The dynamic incremental variant of the ILP approach reduces the long optimization time of the static ILP method with a minor cost penalty compared to a clean-slate offline solution.
Cloud computing has transformed the delivery of modern applications and services by providing scalable, flexible, and cost-effective access to computing resources. One of the most critical challenges in cloud environments is the efficient distribution of dynamic workloads across heterogeneous resources, commonly addressed through load balancing and task scheduling techniques. Efficient scheduling plays a vital role in maximizing resource utilization, minimizing response time, and maintaining acceptable Quality of Service (QoS), particularly under dynamic and large-scale workloads. Despite the progress achieved by traditional heuristics such as Min-Min and metaheuristic approaches like the Improved Sparrow Search Algorithm (ISSA), challenges related to scalability, adaptability, and computational overhead remain. Metaheuristic-based approaches often involve iterative optimization processes that may limit their efficiency in real-time scheduling scenarios. In this paper, we propose a lightweight Stochastic Predictive Energy-Aware Scheduling (SPES) algorithm that integrates predictive execution estimation, multi-resource awareness, and stochastic decision-making. Unlike deterministic scheduling strategies, SPES employs a Top K candidate selection mechanism combined with probabilistic weighting and epsilon-greedy exploration to enhance adaptability and avoid suboptimal resource allocation. The proposed method considers CPU, memory, and I/O demands to achieve balanced utilization across heterogeneous hosts while implicitly addressing energy efficiency through utilization-based modeling. The proposed algorithm is implemented and evaluated using the CloudSim 5.0 simulation framework under heterogeneous multi-region cloud environments with varying workload sizes. Experimental results demonstrate that SPES consistently outperforms ISSA and achieves makespan reductions of up to 23.8% while improving scalability, resource utilization, and scheduling efficiency under dynamic cloud workloads. These results indicate that SPES provides an effective lightweight scheduling solution for large-scale and energy-aware cloud computing environments and supports green computing objectives through improved resource efficiency.
M. Yacoub, A. Raouf, Walaa K. Gad et al.· Electronics· 0 citations
Experiments show that the proposed Hybrid Framework for Joint Optimization of Resource Allocation and Load Balancing that spans two layers in heterogeneous cloud computing systems obtains 25-30% energy savings compared with ordinary methods, significantly reduces p95 latency and also achieves a relatively better Quality Of Service.
Eram Fatma, Nidhi Mishra, Mohammed Abdul Bari· Journal of Intelligent Decis...· 0 citations
A novel QoS-aware task deployment methodology to enhance the Quality of Service (QoS) under resource limitations is introduced and results demonstrate that the proposed method achieves superior system performance compared to existing approaches.
Haotong Zhu, Lei Mo, T. Al-Hasan et al.· ACM Transactions on Design A...· 0 citations
Simulation results confirm that the proposed JORC framework substantially reduces latency, energy consumption, and overall system cost, while increasing the successful task completion ratio compared to existing baseline approaches.
Tanmay Baidya, S. Moh· Italian National Conference...· 0 citations
Partition scheduling in distributed graph computing systems is a key factor affecting overall computational performance. In real-world computing clusters, resource heterogeneity is ubiquitous, and differences in resource capabilities among nodes make it difficult for partition scheduling strategies based on homogeneous assumptions to achieve reasonable load allocation, thereby degrading system execution efficiency. To address this issue, we propose Resource-Aware Scheduling of Graph Partitions (RASGP) for heterogeneous clusters. First, we develop an execution time model for heterogeneous clusters, where the execution time is determined by both computation cost and communication cost. Specifically, the computation cost is characterized by the partition computational load and the comprehensive computing power of a worker, while the communication cost is characterized by the number of cross-node edges and the worker’s bandwidth capability. Based on this execution time model, we design a greedy resource-aware partition scheduling method that achieves more balanced workload distribution by matching partition loads with the resource capacities of different workers. The experimental results demonstrate that RASGP can effectively enhance the graph computing performance in heterogeneous environments under various graph computing tasks and partitioning strategies, particularly in scenarios with a higher degree of heterogeneity. In addition, the scheduling overhead of our method accounts for a small proportion and maintains good controllability as the number of cluster nodes increases.
A task scheduling method using the Deep Q-Network to determine the computation node for the computation task and a dynamic congestion-aware mechanism to determine a low-cost routing path is proposed, which gradually obtains an effective task scheduling scheme through multiple rounds of alternating iterations.