Advances, Systems and Applications
Table 5 Evaluation of Machine Learning Algorithms for Cloud Data Center Resource Management
From: A systematic review on effective energy utilization management strategies in cloud data centers
Author/ Year | Algorithm/Method | Data set/ Workload | Tools/ Experiment Environment | Objective | Performance Metrics/ Pros | Limitations |
|---|---|---|---|---|---|---|
J. Rao et al. [65], 2009 | RLVCONF | TPC | A testbed of cloud with Xen VMs | VM Auto-Configuration, optimize VMs performance | SLA, Throughput, Response Time, Adaptability, and Scalability | Limited samples quality in model training, N/W, and disk B/W not included. |
T. Vinh et al. [66], 2010 | Neural Network | NASA, ClarkNet | CloudSim, GridSim | PM Selection, Load prediction | Energy Consumption, drop rate, predictor | Less diversity of workloads and application services |
O. Niehorster et al. [67], 2011 | SVM | RUBiS | Libvirt 0.8.3 interface, Eucalyptus | Utilization Prediction, Provisioning of VM, SaaS, private cloud | Service Level Objectives (SLO), QoS, self-optimizing, Autonomic Resource Management | Required parallel learning in larger cloud environments, cost estimate misses the SLO, more dataset required. |
G. Kousiouris et al. [55], 2011 | ANN LR | 6 Matlab Benchmark tests | MatlabR2007b | VM performance, VM Analysis | Scheduling decisions, placement of VMs | Detection of workload, Real-world application, premature convergence |
S. Islam et al. [68], 2012 | NN LR | TPC-W | Java | CPU Prediction, resource management, forecasting resource utilization | SLA Fulfilment, Mean Absolute Percentage Error Root Mean Squared Error, R2prediction accuracy, | Need more variety of workload generators, required utility functions for prediction, cost, performance. |
C. Xu et al. [69], 2012 | RL URL | TPC | Real environment | VM Configuration | Service Quality, Throughput, SLA assurance, system utilization | Time complexity, less impact on the quality of final configuration, and traffic perturbations deserve further investigations. |
F. Farahnakian et al. [70], 2013 | DC-KNN | PlanetLab | CloudSim | Utilization Prediction | SLA Violation, Energy Consumption | RAM and N/W resources not included, required K-NN regression for predicting overutilized and under-utilized hosts |
F. Farahnakian et al. [71], 2014 | RL-DC | PlanetLab | CloudSim | Dynamic Consolidation | Energy Consumption, SLA Violation | Real environment required, optimum solution through trial and error in a dynamic context |
M. Patel et al. [72], 2016 | SVR | Real Data set of Xen | R language | Predict Dirty page | Migration Time, Total transferred pages | Model can be overtrained or undertrained, no live migration |
M. Duggan et al. [73], 2016 | AI tech RL RLLM | PlanetLab | CloudSim | Network-aware live VM Migration strategy | Energy Consumption, VM Migration, SLA Violation, PDM, ESV, performance | No real-world cloud applications, |
M. Duggan et al. [74], 2017 | RNN | PlanetLab | CloudSim | Predict CPU utilization | CPU Utilization, Energy Efficiency, Economy of Scale | RAM and disk utilization required, Back propagation through time prediction accuracy |
Q Z Ullah et al. [75], 2017 | ARNN | FastStorage | rJava | CPU Prediction usage | CPU Resource Utilization | Need more duration for prediction, size of training data, type of prediction patterns, temporal and spatial complexity |
R. Shaw et al. [76], 2017 | ARLCA | PlanetLab | CloudSim | Resource management | Energy Consumption, SLA Violation | RAM and N/W bandwidth are not considered, multi-objective optimization techniques are required. |
S. Sotiriadis et al. [77], 2018 | SVM | YCSB | OpenStack | VM Scheduling, VM placement | CPU Utilization, performance, CPU steal time, prediction of VM resource | Need model for classification and regression, time frame window, behavior of VMs and PMs. |
K. Mason et al. [78], 2018 | EvolutionaryNN CMA-ES | PlanetLab | CloudSim | Predict CPU consumption, performance | CPU Utilization, Mean Absolute and Squared Error, Multi-Step Prediction Accuracy | RAM and disk utilization not included, prediction accuracy of a system trained only on the PlanetLab. |
D. Patel et al. [79], 2019 | DT with ANN | Real System | CloudSim, Matlab2015a | VM Migration based Load balancing, performance, and accuracy | Energy Consumption, CPU Utilization, VM Migration | CPU, bandwidth, RAM parameter are not considered, ANN should be integrated with a cloud server for continuous load assessment. |
J. Kumar et al. [80], 2020 | WFNN | Google Cluster Trace, NASA, and Saskatchewan servers’ weblogs | Python3 Jupyter notebook | QoS, Resource Utilization, | Performance, predict upcoming workload with precision, Accuracy, forecast accuracy, faster convergence | Forecast multivariate workload traces, computational complexity, high computation costs |
D. Saxena et al. [81], 2021 | NN | Google cluster dataset | Python version 3 | Resource provisioning, VM Placement, VM Allocation | Performance, Power consumption, QoS, Resource Usage | Network traffic not included, manual selection of nodes in the I/O layer of OM-FNN predictor, more EC due to communication-intensive VMs. |
S. Malik et al. [64], 2022 | FLNN, CNN | Google cluster traces | Â | Multi Resource utilization | Prediction of Resources, Accuracy, Resource Utilization | Predicting disk utilization, cost-effectiveness, and network, multi-variate resource utilization datasets. |