GEESE is a novel UAV based system that enables the dynamic deployment of an edge computing infrastructure through the cooperation of multiple UAVs carrying cloudlets.
- We quantify the performance of UAVs to transport cloudlets to the edge.
- We demonstrate that an edge computing infrastructure can be formed through the cooperation of multiple types of UAVs via collaborative processing.
- We develop an edge delivery model that estimates the amount of UAVs and capabilities of cloudlets required to satisfy a workload from a crowd of users.
- Computing capacity and energy consumption of cloudlets We analyze the computational capacity to handle the task consists of detecting prime numbers within a list of available numbers. Prime number detection Android Web server and the Java client
GEESE delivers cloudlets on demand to any environment to provide edge computing services to end-applications.
Figure 1: A large group of users looking for computation support on the move, a) Using cellular network to reach remote cloud resources, b) Using Wi-Fi/Bluetooth to access constrained edge server in proximity, c) Using the cloudlet support provided by UAVs in proximity.
We implemented a proof-of-concept prototype that follows a master/worker topology. A worker is an idle device, and the master is an initiator device that triggers the execution.
The master is in charge of initiating the task and divide it into jobs based on the number of available devices. The master then sends each job to the workers in a round-robin fashion. Jobs are independent of each other. Thus, the success of the application is measured by the rate of completed jobs. Each worker then processes the job and promptly returns the task to the master once the task is finished. The prototype application is developed in Android version 5.0.1 and implements a Convolutional Neural Network (CNN) model to recognize objects within images. We consider a set of 50 images in a resolution of 224x224, which are collected from the ImageNet We rely on a deep learning implementation that uses TensorFlow with a pre-trained and quantized mobilenet_v1_1.0_224 model for recognizing objects in the images.
Master Android App
Worker Android App including the pre-trained model and the label file
During the experiment, the master node also works as a Wi-Fi hotspot. Workers are connected to the master using Wi-Fi. The workers' application must be executed first, and then the master application. On the master device, make a folder name "TestImages" under the root directory and put the images you want to recognize. You can test the application with this TestImages also. With the new Android versions, you may have to set the app permissions manually before running them.
There is a button on the master application that should be pressed to see the connected workers' IP addresses, and then the master can start sending images to the workers.
This tool is built for conducting experiments to validate our research solutions. If you are using the tool for your research, please do not forget to cite. Thanks!
- Liyanage M, Dar F, Sharma R, Flores H. GEESE: Edge computing enabled by UAVs. Pervasive and Mobile Computing. 2021 Feb 4:101340.
@article{liyanage2021geese,
title={GEESE: Edge computing enabled by UAVs},
author={Liyanage, Mohan and Dar, Farooq and Sharma, Rajesh and Flores, Huber},
journal={Pervasive and Mobile Computing},
volume = {72},
pages={101340},
year={2021},
doi = {https://doi.org/10.1016/j.pmcj.2021.101340},\
publisher={Elsevier}
}