Distributed Decomposed Data Analytics in Fog Enabled IoT Deployments

Mohit Taneja, Nikita Jalodia, Alan Davy

Research output: Contribution to journalArticlepeer-review

33 Citations (Scopus)

Abstract

The edge of the network plays a vital role in an IoT system, serving as an optimal site to perform operation on data before transmitting it over the network. We present the fog specific decomposition of multivariate linear regression as the predictive analytic model in our work using Statistical Query Model and Summation Form. The decomposition method used is not the contribution, but applying the decomposition method to the analytics model to run in a distributed manner in fog enabled IoT deployments is the contribution. What is novel is the decomposition made on a fog based distributed setting. To test the performance, our proposed approach has been applied to a real-world dataset and evaluated using a fog computing testbed. The proposed method avoids sending raw data to the cloud, and offers balanced computation in the infrastructure. The results show an 80% reduction in amount of data transferred to the cloud using the proposed fog based distributed data analytics approach as compared to the conventional cloud based approach. Furthermore, by adopting the proposed distributed approach, we observed a 98% drop in the time taken to arrive to the final result as compared to the cloud centric approach. We also present the results on quality of analytics solution obtained in both approaches, and they suggest that fog based distributed analytics approach can serve as equally as the traditional cloud centric approach.
Original languageEnglish
Article number8675283
Pages (from-to)40969-40981
Number of pages13
JournalIEEE Access
Volume7
Issue number1
DOIs
Publication statusPublished - 2019

Keywords

  • cloud computing
  • data analytics
  • decomposition
  • distributed
  • Fog computing
  • Internet of Things (IoT)

Fingerprint

Dive into the research topics of 'Distributed Decomposed Data Analytics in Fog Enabled IoT Deployments'. Together they form a unique fingerprint.

Cite this