Loading...

Resource Management and Scheduling in Edge-Based Infrastructures

Siar, Hajar | 2023

14 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 56069 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Izadi, Mohammad
  7. Abstract:
  8. Cloud computing has a centralized computing architecture, such that services are provided by cloud servers at far distances from user devices. So, the data gathered by user devices must be sent to cloud servers. This is leading to various challenges, such as bandwidth, latency and privacy, especially in IoT where the number of devices is very large. The devices belong to different users with specific QoS requirements. Edge-based infrastructures or edge paradigms have the potential to tackle these challenges by utilizing devices and servers close to users. This thesis aims to present solutions for resource management problems in edge-based infrastructures, considering the QoS requirements of users. For the first problem, we study task offloading and resource allocation of workflow ensembles. A solution based on hedonic games and stable matching is presented to maximize the number of completed workflows in the ensemble. The evaluation results indicate that using edge/fog and cloud servers yields a 50\% improvement in the ensemble's completed workflows. Then, we narrowed the problem either in terms of architecture and application. In this case, the problem of resource allocation and scheduling of independent and time-sensitive jobs in edge environments is investigated. We formulated this problem using a routing game such that jobs are allocated to resources according to their demands and deadline. We proved the existence of pure Nash equilibrium in this game, then presented a greedy and polynomial-time algorithm called SAFSA to calculate it. Evaluation studies using real-world data-set of Google-cluster confirm the ability of SAFSA to simultaneously minimize the average response time and deadline miss-rate of the jobs. Our approach's performance is robust when there are many IoT devices. By detecting some limitations in simulation tools for resource management and scheduling in edge-based infrastructures and workflow-based applications, we developed a simulation toolkit called WIDESim to overcome some of these limitations. WIDESim can simulate resource management and scheduling of different types of workflow-based applications in distributed systems with a graph topology
  9. Keywords:
  10. Resource Management ; Scheduling ; Cloud Computing ; Game Theory ; Edge-Based Infrastructures ; Internet of Things

 Digital Object List

 Bookmark

No TOC