Skip to main content

Edge and CLoud Computation: A Highly Distributed Software Architecture for Big Data AnalyticS

Periodic Reporting for period 1 - CLASS (Edge and CLoud Computation: A Highly Distributed Software Architecture for Big Data AnalyticS)

Reporting period: 2018-01-01 to 2019-06-30

Can you imagine a city where data is shared between city and cars in real time, to implement intelligent traffic management and advanced driver assistance systems? The CLASS project is working towards sustainable, efficient and safe mobility applications in future smart cities and connected vehicles. In that regard, a key technological challenge is the need of processing large amounts of complex and geographically distributed sources of data coming from vehicles, city infrastructures, IoT devices, etc. This challenge is further complicated with the additional need of dealing with this information in real time conditions.

With a total funding of € 3.9 million, the CLASS project aims to develop a novel software architecture framework to efficiently design, deploy and execute distributed big data analytics workloads capable of processing distributed data sources along the compute continuum (from the edge to the cloud) in a complete and transparent way, while providing real-time guarantees for smart cities, connected cars, and future autonomous vehicles. To do so, the CLASS project defines two strategic goals:
1. To develop a set of tools and methods to provide the right level of abstraction to facilitate the development of complex and distributed big-data workloads, and
2. to increase the real-time capabilities of systems based on big-data analytics, while providing the data throughput speed- up and data analytics accuracy to cope with current increasing trends of volumes, variety and velocity of data.

The CLASS project will be evaluated in the Modena Automotive Smart Area (MASA), a real urban laboratory in the city of Modena, featuring a heavy sensor and device infrastructure to collect data in real-time, and a network connectivity that enables to exchange massive amounts of information. Moreover, CLASS includes three prototypes of connected cars equipped with the necessary connectivity and sensors to give innovative capabilities to drivers.

Overall, the CLASS software architecture will allow to generate a common knowledge base between the city and the cars in real-time and upon which valuable information will be extracted to implement advanced smart city applications, such as intelligent traffic management, air pollution simulation or advanced driving assistance systems. Concretely, CLASS is implementing three particular applications:
- A digital traffic sign application that offers the opportunity to dynamically change traffic conditions based on real-time traffic information,
- a smart parking application that collects real-time information about the available parking places in the monitored area, and
- an obstacle detection application that inform to drivers about general objects and vulnerable road users that may cross the driving path.
During the period covered by this report, the CLASS project has performed two main activities: (1) The development of the CLASS software architecture, and (2) the design of the MASA infrastructure and the three connected cars, upon which the CLASS software architecture will be evaluated.

Regarding the CLASS architecture, the project has carefully selected and developed the set of software components that will form the architecture, prioritising those owned by the members of the CLASS consortium or offered as open-source components with a large community behind them, with the objective of reducing the time-to-market and maximize exploitation opportunities. The CLASS architecture consists of four layers:
- A data analytics layer, which includes a unified development environment capable of exploiting different parallel patterns.
- An orchestration layer capable of efficiently distributing the computation across the compute continuum while providing real-time guarantees.
- An edge and cloud layers capable of efficiently exploiting performance capabilities of edge and cloud computing platforms. At the edge side, the CLASS architecture features the most advanced parallel embedded architectures; at the cloud side, it features a Container as a Service solution.

The CLASS software components are being developed with well-defined interfaces to facilitate the interoperability, portability, and scalability properties needed to provide developers with a solid infrastructure for better exploiting the kind of distributed environments where the CLASS project is built upon.

Regarding the MASA, it has been wired with an optical fiber network connected to the Metropolitan Area Network of the City of Modena, and 4.5G 5G and Lo-Ra antennas have been installed. This communication infrastructure allows to collect data from multiple IoT devices, including 12 cameras connected to the city computing nodes and 6 traffic light counters. Finally, the MASA includes four powerful edge nodes based on NVIDIA GPUs, and a data-center in which the CLASS cloud layer executes.
Regarding the connected cars, CLASS has installed multiple sensors in two Maserati vehicles, a Quattroporte and a Levante, including 6 cameras with different field of view, a GPS/GNSS for a determining the vehicle location and a LiDAR. 4G-LTE antenna receivers have been also installed to transmit information captured and processed by the vehicles. Finally a computing platform based on NVIDIA GPUs have been installed.
The existing big-data systems are designed to provide either quick and reactive responses in real time (data-in-motion) or thorough and more computationally intensive feedback based on the analysis of a vast amount of information (data-at-rest). Although these options have been tackled separately, combining the two provides complementary capabilities and benefits.

On the edge computing side, the availability of new parallel heterogeneous embedded processor architectures, (e.g. GPUs, many-core fabrics, FPGAs) may enable the processing of complex data-in-motion analytics with a reduced pressure on communication and cost within vehicles or in cabinets installed across the city. On the cloud computing side, data-in-motion analysis can be completed (and enriched) with data-at-rest analysis with a more complex big-data workloads. The vision of CLASS is that the pressure that the newest smart systems requiring big data analytics and real-time requirements will put on computing systems, can be efficiently addressed by combining data-in-motion and data-at-rest analytics into a single computing workflow that can be efficiently distributed across the compute continuum, from cloud in which data-servers are located, to data sources and edge devices that are co-located with them.

CLASS has developed a novel software component capable of optimally allocating and efficiently distributing the different analytics methods that form an heterogeneous big-data analytics workflow across the compute continuum. Our allocation strategy selects those computing platforms in which execution time and the data transfer time is minimized, allowing to reduce the execution time variability and so providing better real-time guarantees.

This allocation capabilities are aimed to be applied to the CLASS use-case, which we aim to generate several beneficial outcomes by a more efficient use of computing and communication resources, including:
- Improving overall traffic management by 20%
- Reducing pollution by 20%
- Reducing the response time of emergency vehicles by 30%
- Reducing the number of accidents by 30%
- Reducing the time spent looking for a parking space by 40%
The Modena Automotive Smart Area (MASA)
knowledge base shared between the car and the city
Description of the compute continuum, including car and city computing resources
Connected vehicle with the LiDAR