Cloud-Based Data Centre with Predictive Machine Learning Server Traces

Citation Author(s):
Niladri
Dey
KL University
Gunasekhar
T
KL University
Submitted by:
Niladri Dey
Last updated:
Tue, 08/11/2020 - 01:43
DOI:
10.21227/93kc-4251
License:
889 Views
Categories:
5
1 rating - Please login to submit your rating.

Abstract 

Cloud Computing is been the field for interest by the research community and application development industry for few decades now. The ease of development, deployment and management of applications from wide range of computing paradigm and ability to manage the applications over network enabled systems are the biggest selling points of cloud computing. These benefits are materialized using the mechanism called virtualization on cloud computing and in cloud-based Data Centres. The virtualization technology not only enables the complete virtual view of the physical resource pools, rather also enables few key benefits such as portability, recovery after failure and most prominently generates the fundamental technique for load balancing. In the recent past, a good number of research outcomes are observed in improving the load balancing mechanisms using stochastic methods or by deploying bio-genetic optimization methods. Nonetheless, demand for higher scalability and higher availability for the applications are the constant catalyst for further exploration betterment of the load balancing algorithms. In this work, the fundamental problems with load balancing are analyzed along with the drawbacks of the existing load balancing genetic optimization methods. Primarily the genetic optimization methods are criticized for higher order probabilistic distributions, higher complexity of the solution search spaces and the increasing computational complexity while introducing additional parameters for improving the decision-making capabilities. Thus firstly, this work addresses the possibilities of load characteristics summarization for better implementation of the load prediction analogies using the service request type categorizations. Secondly, for the load summarization process, the equivalence of the related parameters responsible for load calculations are reduced to abridged parametric representations using the correlation-based reduction strategy, which significantly supports of the principle of load characteristics summarization process and finally, this work demonstrates anther strategy for pheromone level prediction using the corrective coefficient based-based method for enhanced prediction over the other standard genetic optimization methods during load balancing. This work utilizes the mathematical modelling methods, data representation methods, simulation over multiple simulation tools and finally the web-based third-party cloud service providers to test and prove the concepts proposed in this work, which are furnished in the subsequent sections of this work. The resultant approach is proven to be the most balanced strategy for power consumption, less VM migrations, reduced time complexity and finally less violation of the service level agreements compared with the other parallel standard methods

Instructions: 

The files are in text format and any programming language FileIO methods can be used. 

No special requirements to read the files.