Runtime-Aware Hybrid Plan-based and Backfilling Algorithm for Enhancing Healthcare Task Scheduling in Cloud-IoT Environment

Cloud computing and the Internet of Things (Cloud-IoT) in medical applications offer a promising avenue for enhancing healthcare data processing and analysis.Task Scheduling (TS) in Cloud-IoT is a vital challenge for healthcare services, impacting user fulfilment time.To combat this, several bio-inspired schemes were applied to address this issue, including the Hybrid Particle Swarm Optimization and Salp Swarm Algorithm (HPSOSSA).While HPSOSSA effectively reduces the waiting period, computation cost, makespan, and execution period of clinical demands on Cloud-IoT, it primarily focuses on minimizing CPU usage of VMs and maximizing medical resource utilization.However, optimizing other parameters such as throughput, memory utilization, and bandwidth is equally important.This article introduces a new hybrid TS algorithm to consider throughput, memory, and bandwidth requests for efficient TS.First, a Quasi-Recurrent Neural Network (Quasi-RNN) is used as a meta-learner to estimate Task Runtime Reliability (TRR), categorizing tasks into predictable and unpredictable tasks.Tasks with high Expected Runtime Accuracy (ERA) are scheduled utilizing a plan-based method with the Walrus Optimization Algorithm (WOA) while the residual ones are backfilled using the Deviation Backfilling (DB).To optimize resource utilization, a portion of CPU hubs is saved for backfilling and adjusted adaptively according to the Resource Use Ratio (RUR) of predictable tasks among newly submitted tasks.The proposed algorithm is called Hybrid Quasi-RNN with Walrus optimizer and DB (HQRWDB).The performance of this HQRWDB algorithm for TS in the Cloud-IoT system is evaluated using an HCSP-based dataset with scenario high heterogeneity of cloudlet with low heterogeneity of VMs.Simulation outcomes reveal that the HQRWDB achieves a 1943sec makespan, 0.9 average RUR, 0.09 Degree of Imbalance (DoI), 0.66 tasks/sec throughput, 37MB memory usage, and 224MB/sec bandwidth usage for 1000 cloudlets, compared to the HPSOSSA, Multi-Objective TS using an improved Ant Colony Optimization (MOTS-ACO), PSO with Phagocytosis-based Genetic Algorithm (PSO-PGA) and Multi-objective Grey Wolf Optimization (MGWO) algorithms for TS in Cloud-IoT.