A GRU-Based Prediction Framework for Intelligent Resource Management at Cloud Data Centres in the Age of 5G

Abstract

The increasing deployments of 5G mobile communication system is expected to bring more processing power and storage supplements to Internet of Things (IoT) and mobile devices. It is foreseeable the billions of devices will be connected and it is extremely likely that these devices receive compute supplements from Clouds and upload data to the back-end datacentres for execution. Increasing number of workloads at the Cloud datacentres demand better and efficient strategies of resource management in such a way to boost the socio-economic benefits of the service providers. To this end, this paper proposes an intelligent prediction framework named IGRU-SD (Improved Gated Recurrent Unit with Stragglers Detection) based on stateof-art data analytics and Artificial Intelligence (AI) techniques, aimed at predicting the anticipated level of resource requests over a period of time into the future. Our proposed prediction framework exploits an improved GRU neural network integrated with a resource straggler detection module to classify tasks based on their resource intensity, and further predicts the expected level of resource requests. Performance evaluations conducted on realworld Cloud trace logs demonstrate that the proposed IGRU-SD prediction framework outperforms the existing predicting models based on ARIMA, RNN and LSTM in terms of the achieved prediction accuracy.

START: Straggler Prediction and Mitigation for Cloud Computing Environments using Encoder LSTM Networks

“…Within this category, other techniques use prediction models to apriori determine the set of tasks in a job that might be stragglers. Examples include RPPS [23] and IGRU-SD [22]. When considering mitigation, approaches either avoid straggler tasks or prevent high response times by methods such as rescheduling, balancing load or running job replicas (clones).…”

“…When considering mitigation, approaches either avoid straggler tasks or prevent high response times by methods such as rescheduling, balancing load or running job replicas (clones). Examples of such strategies include Dolly [20], GRASS [8], SMT [28] SMA [14] RDD [19] Mitigation Only Methods LATE [29] Dolly [20] GRASS [8] Dolly [20] GRASS [8] Wrangler [17] Prediction based Mitigation Methods SGC [9] IGRU-SD [22] START (this work)…”

“…Here, straggler prediction means the prediction of straggler tasks before execution. In particular, [22], [23] use deep learning based solutions to predict straggler tasks and efficiently manage them.…”

“…Specifically, diversity in task execution times leads to the presence of tasks with extremely high or low execution times. This makes the state space of the neural network very large when modelling the distribution of task response times and hence it is often omitted in practical approaches [22], [23]. Second, these approaches ignore the heterogeneous host capabilities, which can also lead to poor scheduling or mitigation decisions [21].…”

“…Prediction allows early mitigation, reducing the SLA violation rate and execution time and maintaining QoS at the required level. Our performance evaluation is carried out using CloudSim 5.0 [27] and compares our technique with well-known existing techniques (SGC [9], Dolly [20], GRASS [8], NearestFit [6], Wrangler [17], and IGRU-SD [22]) in terms of QoS parameters such as energy consumption, execution time, resource contention, CPU utilization and SLA violation rate. Experimental results demonstrate that START gives lower execution time and SLA violations than existing techniques, also offering low computational overhead.…”