Latency-Aware Edge-Cloud Collaboration for Real-Time Video Analytics

Bidkar, Yash Santosh (2025) Latency-Aware Edge-Cloud Collaboration for Real-Time Video Analytics. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (475kB) | Preview

Abstract

The growth of real-time video analytics solutions through IoT has created a large amount of network load for traditional cloud architectures. As a result, the development and implementation of edge computing systems to allow for the processing of video data at or near the source has become necessary. In addition, fixed allocations of analytics workflows are not efficient for rapidly changing networks where data may move between devices at very different rates. This paper presents a framework to collaboratively manage the scheduling of video analytics workflows across an edge device and a cloud computing service provider. A key component of this framework is a predictive scheduling algorithm based on the ARIMA model to predict the time delays for data when transferring from one device to another. By being able to accurately predict the delay, decisions regarding where to offload the task can be made to reduce real-time pressure. A series of experiments using a real worldwide data network testbed with a local edge device and AWS EC2 instance provide empirical results to validate the proposed system. Scheduling strategies using reactive and predictive schedulers were compared to a cloud-only baseline. The comparison showed that the cloud-only context experienced a critical failure when the high-latency environment existed where it had an extremely high deadline miss rate of 98.89%. The reactive and predictive schedulers, however, met the real time deadlines for each frame they were processing by avoiding the high-latency cloud path due to their intelligent scheduling algorithm. The performance of the predictive scheduler was also better than that of the reactive scheduler in terms of frames per second throughput, achieving a maximum throughput of 2.36 FPS while the reactive scheduler only had a maximum throughput of 2.00 FPS. Clearly, the ability to use forecasts as the basis for proactive decisions provides substantial benefit to the performance of a scheduling system.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Heeney, Sean UNSPECIFIED
Subjects:	T Technology > T Technology (General) > Information Technology > Cloud computing Q Science > QA Mathematics > Electronic computers. Computer science > Computer Systems > Computers > Electronic data processing > Electronic data processing--Distributed processing > Edge computing T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science > Computer Systems > Computers > Electronic data processing > Electronic data processing--Distributed processing > Edge computing T Technology > TK Electrical engineering. Electronics. Nuclear engineering > Telecommunications > Computer networks > Internet of things
Divisions:	School of Computing > Master of Science in Cloud Computing
Depositing User:	Ciara O'Brien
Date Deposited:	31 Aug 2026 12:23
Last Modified:	31 Aug 2026 12:23
URI:	https://norma.ncirl.ie/id/eprint/9691

Actions (login required)

View Item