Project Coordinator (EU) :Department of Information Engineering - University of Pisa
Country of the EU Coordinator :Italy
Organisation Type :Academia
Project participants :
Carlo Vallati is an Associate Professor at the Department of Information Engineering of the University of Pisa
Francesca Righetti is a Postdoctoral Research Fellow at the Information Engineering Department at the University of Pisa.
Giuseppe Anastasi is a Full Professor of computer engineering at the Department of Information Engineering of the University of Pisa.
Nirmalya Roy is an Associate Professor in the Information Systems department at University of Maryland Baltimore County.
Aryya Gangopadhyay is a Professor and the chair of the Information Systems Department at the University of Maryland Baltimore County.
Bipendra Basnyat, P.E., is a final-year Ph.D. student at the University of Maryland Baltimore County.
State of US partner :Maryland
Starting date :
EdgeFlooding: Exploiting Edge computing for Real-Time Monitoring and Detection of Flash Floods
The experiments, carried out by UNIPI, aim at verifying the feasibility of extending the flood monitoring system designed at UMBC, based on a centralized cloud computing approach (Figure 1), by adopting an edge/cloud computing approach, thus distributing some of the data analysis functions at the edge (Figure 2). As one of the first research efforts in the area of government environment monitoring, the decentralization of the functionalities in the implementation of the flood monitoring system is expected to have significant impact and specifically:
- improve system scalability and ensure cost efficacy, as functions decentralization can ensure large-scale scalability by design;
- minimization of data transmission and reduction of the load on the cloud reduce costs, which is expected to be crucial to foster the diffusion of such system worldwide;
- improve system reliability as a decentralized implementation that does not rely on continuous data offloading and data transmission can mitigate the consequences of network outages;
- reduce latency through direct communication between the monitoring stations and the service analysing the data, thus increasing system responsiveness.
Implementation plan :
In order to perform a comprehensive evaluation of the flood monitoring system and to have a fair term of comparison to assess the feasibility of both the cloud and edge computing approaches in a real deployment, we define two configurations for our experiments. Specifically, we define the cloud configuration and the edge configuration. The experiments run with the cloud configuration will allow us to have a performance baseline and subsequently evaluate the performance gap/gain between the two approaches.
The collaboration between UNIPI and UMBC teams, fostered by their complementary background and experience, will pave the way for a long-term collaboration that will contribute to enrich the EU – US ecosystem of top researcher. The project will establish the opportunity for joint research activities not only in the area of novel environmental monitoring systems but also in the broader area of smart computing systems. This will open the possibility for other joint project proposals to be submitted to other EU-US joint calls.
The opportunity of establishing a long-term collaboration will be nurtured during the EdgeFlooding activities and also after the end of the project by assessing the opportunity to perform the following:
- establish a permanent visiting program between the two research groups;
- sign a memorandum of understanding between UNIPI and UMBC in order to foster a broader collaboration that involves different research groups, beyond the two teams involved in this proposal.
The results of the experiments also have the potential of having an impact on both US – EU solutions for flood monitoring. Existing environmental monitoring systems are available today, however, they adopt a centralized cloud-based approach. Although this approach is widely adopted today as it can guarantee low-cost and rapid development, future large-scale implementations of such systems will require the adoption of a decentralized architecture. With this respect, the experiment is expected to provide metrics and measurements on the performance of flood monitoring systems that adopt a distributed edge computing approach. Such results will provide meaningful insights to assess the feasibility of adopting this approach for the implementation of real systems. In addition to this, such results will provide meaningful data that could be considered also in other environmental monitoring systems, in particular the ones that leverage sensor data and image analysis. The experiment and the analysis of the results will offer a general indication on the performance of such systems, providing an indication on the possibility of adopting a distributed approach also in other contexts.
The main goal of the experiment is to assess the feasibility of adopting a distributed approach for the implementation of the flood monitoring system by comparing its performance with the centralized approach where the data analysis process is hosted on powerful servers on the Cloud.
The KPI's metrics will provide an insight to assess if the distributed approach can support an efficient execution of the flood monitoring system as a whole and how many resources the system consume on the infrastructure. Those results are expected to offer the following results:
- Identify the hardware configuration that can support the implementation of a flood monitoring system at the edge, i.e. the subset of hardware configurations, among the ones considered in our experiments, that can ensure sufficient performance for the video image analysis process when executed at the edge.
- Measure the gap/loss in performance when the video image analysis is performed at the edge, with less resources in terms of CPU, RAM and GPU capabilities, with respect to the Cloud based approach.
- Measure the gain in terms of communication bandwidth consumption when the video feed is not transmitted to the Cloud over the Internet but it is analysed on a system in close proximity. - Measure the amount of resources consumed in both the approaches by the system to assess its scalability.
Those results will be analysed at the end of the experiments, their analysis is expected to provide guidelines for the implementation of real systems.