FELARE: Fair Scheduling of Machine Learning Tasks on Heterogeneous Edge Systems

Abstract

Edge computing enables smart IoT-based systems via concurrent and continuous execution of latency-sensitive machine learning (ML) applications. These edge-based machine learning systems are often battery-powered (i.e., energy-limited). They use heterogeneous resources with diverse computing performance (e.g., CPU, GPU, and/or FPGA) to fulfill the latency constraints of ML applications. The challenge is to allocate user requests for different ML applications on the Heterogeneous Edge Computing Systems (HEC) with respect to both the energy and latency constraints of these systems. To this end, we study and analyze resource allocation solutions that can increase the on-time task completion rate while considering the energy constraint. Importantly, we investigate edge-friendly (lightweight) multiobjective mapping heuristics that do not become biased toward a particular application type to achieve the objectives; instead, the heuristics consider "fairness" across the concurrent ML applications in their mapping decisions. Performance evaluations demonstrate that the proposed heuristic outperforms widely-used heuristics in heterogeneous systems in terms of the latency and energy objectives, particularly, at low to moderate request arrival rates. We observed 8.9% improvement in on-time task completion rate and 12.6% in energy-saving without imposing any significant overhead on the edge system.

A Novel Fault-Tolerant Aware Task Scheduler Using Deep Reinforcement Learning in Cloud Computing

“…Methodology Parameters [17] RL to make a scheduling model for cloud computing Makespan, total power cost in datacenters, energy consumption, migration time [18] RL task scheduling with Q-learning Makespan, total power cost, energy consumption [19] Task scheduling optimization algorithm called MPSO Makespan, energy consumption, packet delivery ratio, trust value [20] FELARE Ontime task completion rate, energy saving [21] RLFTWS Makespan, resource usage [22] WBCNF Computation time of tasks [23] BRCH-GWO Makespan [24] RELIEF Communication delay, reliability [25] PSO-based multipurpose algorithm Turnaround time, makespan [26] Regressive WO algorithm Processing cost, load balancing tasks [27] GO Makespan, resource utilization [28] Dynamic task scheduling algorithm based on an improved GA Total execution time and resource utilization ratio [29] PSO based on an AC algorithm Task completion time, makespan [30] DRL-based task scheduling Makespan, computation time [31] MRLCC is an approach for organizing tasks that are based on Meta RL Energy consumption, total cost, makespan [32] A novel DRL-based framework Cost and throughput, makespan [33] RLFTWS Execution time, degree of imbalance [15] DRL model Response time, makespan, CPU utilization [34] Deep reinforcement learning with PPSO SLA violation, makespan [35] DRL Cloud is an NDR-learning-based RP and TS system Estimated completion time, resource utilization [36] Deep Q-network model Degree of imbalance, cost, makespan [37] SDM reinforcement learning Energy consumption, resource utilization [38] DRLHCE Response time, degree of imbalance [39] DQN Makespan, total cost [40] Reinforcement learning Makespan [41] DDDQN-TS Task response time [31] Q-learning Makespan…”