Journal of Cloud Computing: Advances, Systems and Applications (Dec 2022)

Improving cloud efficiency through optimized resource allocation technique for load balancing using LSTM machine learning algorithm

  • Moses Ashawa,
  • Oyakhire Douglas,
  • Jude Osamor,
  • Riley Jackie

DOI
https://doi.org/10.1186/s13677-022-00362-x
Journal volume & issue
Vol. 11, no. 1
pp. 1 – 17

Abstract

Read online

Abstract Allocating resources is crucial in large-scale distributed computing, as networks of computers tackle difficult optimization problems. Within the scope of this discussion, the objective of resource allocation is to achieve maximum overall computing efficiency or throughput. Cloud computing is not the same as grid computing, which is a version of distributed computing in which physically separate clusters are networked and made accessible to the public. Because of the wide variety of application workloads, allocating multiple virtualized information and communication technology resources within a cloud computing paradigm can be a problematic challenge. This research focused on the implementation of an application of the LSTM algorithm which provided an intuitive dynamic resource allocation system that analyses the heuristics application resource utilization to ascertain the best extra resource to provide for that application. The software solution was simulated in near real-time, and the resources allocated by the trained LSTM model. There was a discussion on the benefits of integrating these with dynamic routing algorithms, designed specifically for cloud data centre traffic. Both Long-Short Term Memory and Monte Carlo Tree Search have been investigated, and their various efficiencies have been compared with one another. Consistent traffic patterns throughout the simulation were shown to improve MCTS performance. A situation like this is usually impossible to put into practice due to the rapidity with which traffic patterns can shift. On the other hand, it was verified that by employing LSTM, this problem could be solved, and an acceptable SLA was achieved. The proposed model is compared with other load balancing techniques for the optimization of resource allocation. Based on the result, the proposed model shows the accuracy rate is enhanced by approximately 10–15% as compared with other models. The result of the proposed model reduces the error percent rate of the traffic load average request blocking probability by approximately 9.5–10.2% as compared to other different models. This means that the proposed technique improves network usage by taking less amount of time due, to memory, and central processing unit due to a good predictive approach compared to other models. In future research, we implement cloud data centre employing various heuristics and machine learning approaches for load balancing of energy cloud using firefly algorithms.

Keywords