EdgeDrones: Co-scheduling of drones for multi-location aerial computing missions
Authors: Awada, U., Zhang, J., Chen, S., Li, S. and Yang, S.
Journal: Journal of Network and Computer Applications
Volume: 215
eISSN: 1095-8592
ISSN: 1084-8045
DOI: 10.1016/j.jnca.2023.103632
Abstract:Low altitude platform (LAP) unmanned aerial vehicles (UAVs), also called drones, are currently being exploited by Edge computing (EC) systems to execute complex resource-hungry use cases, such as virtual reality, smart cities, autonomous vehicles, etc., by attaching portable edge devices on them. However, a typical drone has limited flight time, coupled with the resource-constrained attached edge device, which can jeopardize aerial computing missions if they are not holistically taking into consideration. Moreover, the fundamental challenge is how to co-schedule multi-drone among multi-location where EC services are needed, such that drones are scheduled to maximize the utility from the activities while meeting computing resource and flight time constraints. Therefore, for a given fleet of drones and tasks across disjointed target locations in a city, we derive a machine learning (ML) linear regression model that estimates these tasks resource requirement and execution time. Leveraging this estimation values, we jointly consider each drone's flight time availability and its attached edge device resource capacity, and formulate a novel Multi-Location Capacitated Mission Scheduling Problem (MLCMSP) that selects suitable drones and co-schedules their flight routes with the least total distance to visit and execute tasks at the target locations. Then, we show that faster scheduling and execution of complex tasks at each location, while considering the inter-task dependencies is important to achieve effective solution for our MLCMSP. Hence, we further propose EdgeDrones, a variant bin-packing optimization approach through gang-scheduling of inter-dependent tasks that co-schedules and co-locates tasks tightly so as to achieve faster execution time, as well as to fully utilize available resources. Extensive experiments on Alibaba cluster trace with information on task dependencies (about 12,207,703 dependencies) show that EdgeDrones achieves up to 73% higher resource utilization, up to 17.6 times faster executions, and up to 2.87 times faster flight travel time compared to the baseline approaches.
https://eprints.bournemouth.ac.uk/38393/
Source: Scopus
EdgeDrones: Co-scheduling of drones for multi-location aerial computing missions
Authors: Awada, U., Zhang, J., Chen, S., Li, S. and Yang, S.
Journal: JOURNAL OF NETWORK AND COMPUTER APPLICATIONS
Volume: 215
eISSN: 1095-8592
ISSN: 1084-8045
DOI: 10.1016/j.jnca.2023.103632
https://eprints.bournemouth.ac.uk/38393/
Source: Web of Science (Lite)
EdgeDrones: Co-Scheduling of Drones for Multi-Location Aerial Computing Missions
Authors: Awada, U., Zhang, J., Chen, S., Li, S. and Yang, S.
Journal: Journal of Network and Computer Applications
Publisher: Elsevier
ISSN: 1084-8045
https://eprints.bournemouth.ac.uk/38393/
Source: Manual
EdgeDrones: Co-Scheduling of Drones for Multi-Location Aerial Computing Missions
Authors: Awada, U., Zhang, J., Chen, S., Li, S. and Yang, S.
Journal: Journal of Network and Computer Applications
Volume: 215
Publisher: Elsevier
ISSN: 1084-8045
Abstract:Low altitude platform (LAP) unmanned aerial vehicles (UAVs), also called drones, are currently being exploited by Edge computing (EC) systems to execute complex resource-hungry use cases, such as virtual reality, smart cities, autonomous vehicles, etc., by attaching portable edge devices on them. However, a typical drone has limited flight time, coupled with the resource-constrained attached edge device, which can jeopardize aerial computing missions if they are not holistically taking into consideration. Moreover, the fundamental challenge is how to co-schedule multi-drone among multi-location where EC services are needed, such that drones are scheduled to maximize the utility from the activities while meeting computing resource and flight time constraints. Therefore, for a given fleet of drones and tasks across disjointed target locations in a city, we derive a machine learning (ML) linear regression model that estimates these tasks resource requirement and excution time. Leveraging this estimation values, we jointly consider each drone’s flight time availability and its attached edge device resource capacity, and formulate a novel Multi-Location Capacitated Mission Scheduling Problem (MLCMSP) that selects suitable drones and co-schedules their flight routes with the least total distance to visit and execute tasks at the target locations. Then, we show that faster scheduling and execution of complex tasks at each location, while considering the inter-task dependencies is important to achieve effective solution for our MLCMSP. Hence, we further propose EdgeDrones, a variant bin-packing optimization approach through gang-scheduling of inter-dependent tasks that co-schedules and co-locates tasks tightly so as to achieve faster execution time, as well as to fully utilize available resources. Extensive experiments on Alibaba cluster trace with information on task dependencies (about 12,207,703 dependencies) show that EdgeDrones achieves up to 73% higher resource utilization, up to 17.6 times faster executions, and up to 2.87 times faster flight travel time compared to the baseline approaches.
https://eprints.bournemouth.ac.uk/38393/
Source: BURO EPrints