2021/11/19 by Shreshth Tuli, Tuli, Shreshth, Sukhpal Singh Gill +9 · 1 citation
Computer Science · #Cloud Computing and Resource Management #Distributed #FOS: Computer and information sciences #IoT and Edge/Fog Computing #Parallel #Performance (cs.PF) #Software System Performance and Reliability #and Cluster Computing (cs.DC)
paper · pdf · doi:10.48550/arxiv.2111.10241
openalex publication_date 2021/11/19 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Modern large-scale computing systems distribute jobs into multiple smaller\ntasks which execute in parallel to accelerate job completion rates and reduce\nenergy consumption. However, a common performance problem in such systems is\ndealing with straggler tasks that are slow running instances that increase the\noverall response time. Such tasks can significantly impact the system's Quality\nof Service (QoS) and the Service Level Agreements (SLA). To combat this issue,\nthere is a need for automatic straggler detection and mitigation mechanisms\nthat execute jobs without violating the SLA. Prior work typically builds\nreactive models that focus first on detection and then mitigation of straggler\ntasks, which leads to delays. Other works use prediction based proactive\nmechanisms, but ignore heterogeneous host or volatile task characteristics. In\nthis paper, we propose a Straggler Prediction and Mitigation Technique (START)\nthat is able to predict which tasks might be stragglers and dynamically adapt\nscheduling to achieve lower response times. Our technique analyzes all tasks\nand hosts based on compute and network resource consumption using an Encoder\nLong-Short-Term-Memory (LSTM) network. The output of this network is then used\nto predict and mitigate expected straggler tasks. This reduces the SLA\nviolation rate and execution time without compromising QoS. Specifically, we\nuse the CloudSim toolkit to simulate START in a cloud environment and compare\nit with state-of-the-art techniques (IGRU-SD, SGC, Dolly, GRASS, NearestFit and\nWrangler) in terms of QoS parameters such as energy consumption, execution\ntime, resource contention, CPU utilization and SLA violation rate. Experiments\nshow that START reduces execution time, resource contention, energy and SLA\nviolations by 13%, 11%, 16% and 19%, respectively, compared to the\nstate-of-the-art approaches.\n