Page 43 - Kaleidoscope Academic Conference Proceedings 2021

DEVICELESS: A SERVERLESS APPROACH FOR THE INTERNET OF THINGS

Zakaria Benomar ; Francesco Longo ; Giovanni Merlino ; Antonio Puliafito 1,2
1,2
1,2
1,2
1
Department of Engineering, University of Messina, Italy
2
CINI: National Interuniversity Consortium for Informatics, Rome, Italy

ABSTRACT IoT applications on a larger scale since the capital expenditure
of IoT infrastructure is often non-trivial. Besides, getting the
Developers of cloud-native applications have rapidly adopted authorization to install IoT nodes in public domains (e.g.,
the Serverless/Function-as-a-Service (FaaS) paradigm as smart city scenario) is not (always) an easy task. Such an
it exempts them from provisioning and operating the aspect is not a problem in private domains/campuses, but it
infrastructure. Within this context, an interesting approach would become a significant hurdle to overcome in wide and
that can foster IoT applications’ development is extending the large-scale deployments.
serverless paradigm towards the network edge to cover IoT
environments: a paradigm that we refer to as ‘deviceless’. On the other hand, with the rise of cloud computing, we
In this approach, an IoT infrastructure composed of have seen a transition from buying and managing bare metal
devices deployed at the network edge can seamlessly be servers to using instances (in the form of virtual machines
integrated as an application execution infrastructure to or containers) hosted in a cloud data center. Recently,
enable interactions with hosted sensors/actuators. This paper we have even seen a shift from cloud-based instances to
discusses several perspectives available in the literature the serverless computing model [3] where all traces of the
on the (IoT) edge-based serverless paradigm and related actual server platform have disappeared. Specifically, the
use cases in an IoT/edge computing context. Besides, application developer still writes the server-side logic, but,
we present our preliminary prototype for implementing unlike traditional architectures, it runs on stateless compute
the deviceless approach to show its viability. We containers that are event-triggered, ephemeral (may only last
exploit the deviceless paradigm to conceive data pipelines for one invocation), and fully managed by a third party. The
under a flow-based development environment leveraging a developers can focus then only on the business logic of their
geographically distributed IoT infrastructure. applications whilst delegating all infrastructure management
Keywords - Cloud computing, deviceless, edge computing, tasks (i.e., scalability, provisioning, etc.) to the cloud
FaaS, IoT, serverless, virtualization. provider, thus leading to a new utility computing scheme,
namely Function-as-a-Service (FaaS).
1. INTRODUCTION
In the context of enabling seamless interactions with IoT
Over the last few years, industry and academic research resources (e.g., sensors and actuators), our approach aims at
communities have proposed many Internet of Things (IoT) extending the cloud serverless paradigm towards the network
applications. To deal with IoT data management and edge to use it on top of a (shared) IoT infrastructure.
processing, most of the solutions rely on cloud platforms [1]. The applications’ developers can then make use of the IoT
However, these cloud-oriented solutions, usually adopted resources (i.e., sensors and actuators) in a serverless-like
in sensing resource management, can be framed into the fashion without managing the infrastructure or the used
data-centric category [2] as the only operations provided communication protocols. We refer to this new computing
are data manipulation ones. Albeit the large amount of model as deviceless. It is worth mentioning here that the
resources the cloud offers (e.g., compute and storage), cloud objective of the approach is to simplify the way of conceiving
platforms often consider the IoT devices as only data providers IoT applications: the software engineering aspect.
uploading data towards data centers. The cloud role is then
restricted to a scalable sink dealing with data processing. In this paper, we introduce our deviceless approach to explore
This management approach has several drawbacks stemming the serverless concept in an IoT scenario. Its implementation
from the non-real-time access to IoT data and the inability is based on our Stack4Things (S4T) framework and two
to personalize the business logic running on the IoT nodes. OpenStack-based subsystems, Qinling and Zun, that have
To have more flexibility, a user can opt for setting up its been customized for this purpose. Analogously with how the
own IoT infrastructure and interacting with it through ad-hoc serverless paradigm exempts users from managing/operating
and vendor lock-in kind of application-level Application the infrastructure, the S4T deviceless abstraction model
Programming Interfaces (APIs). Nevertheless, the necessity dispenses developers from managing the IoT infrastructure
in such a kind of IoT deployment model to manage and while dealing with the typical constraints deriving from
especially own the infrastructure is a limitation for adopting typical IoT deployments such as Network Address Translators

– xxxix –