Using tiny-ML models to QoS-aware workload distribution at the network edge

Using tiny-ML models to QoS-aware workload distribution at the network edge


Using tiny-ML models to QoS-aware workload distribution at the network edge

نوع: Type: thesis

مقطع: Segment: PHD

عنوان: Title: Using tiny-ML models to QoS-aware workload distribution at the network edge

ارائه دهنده: Provider: Mohammadreza pourhoseini

اساتید راهنما: Supervisors: Dr.Mahdi abbasi

اساتید مشاور: Advisory Professors:

اساتید ممتحن یا داور: Examining professors or referees: Dr.Mirhossein dezfolian-Dr.Hatam abdoli

زمان و تاریخ ارائه: Time and date of presentation: 2023/09/19 16:00

مکان ارائه: Place of presentation: Faculty of Engineering - Class 23

چکیده: Abstract: The number of devices connected to the Internet of Things network has been expanded rapidly. This issue has caused a significant increase in the computational load in the network. To overcome this challenge, cloud computing was presented as a suitable solution. However, cloud computing suffers significant delay to process workloads. Processing workloads at the edge of the network, in addition to reducing the response time, increases the quality of service. Also, the resource limitation at the edge of the network should be taken into account. Therefore, in addition to distributing the workloads at the edge of the network and maintaining the balance between energy consumption and delay, the limitation of resources such as memory consumption should be considered. In this paper, an online method based on XCS learning classifier systems (LCS), named TinyXCS, and an offline method based on decision tree, named TinyDT, are proposed to balance energy consumption and reduce delay in processing workloads considering the memory limitation at the edge of the network. The results of our experiments show the superiority of TinyXCS and TinyDT over similar methods. The simulation shows that in addition to workload distribution, the proposed methods can simultaneously reduce delay and energy consumption and create a compromise between them and memory consumption

فایل: ّFile: