You are on page 1of 6

Management, orchestration and workflow automation

of Edge Computing services: The TANDEM


approach

Konstantinos Vasilopoulos, Vasileios Theodorou, Theodoros Bozios,


Maria-Evgenia Xezonaki and Dimitrios Laskaratos
Emails: {tmpo, konvasilop, maxez, dlaskaratos, theovas}@intracom-telecom.com
Intracom SA Telecom Solutions

Abstract—The distributed computing paradigm that brings Since infrastructure capacity at the network edge is usu-
information processing closer to end users and data sources, ally limited compared to centralized cloud resources, smart
i.e., Edge Computing, is growing in popularity and adoption in mechanisms are required for the efficient resource manage-
many different domains, such as Cloud Computing, Telecommu- ment and distribution of services between edge nodes and
nications, Internet of Things (IoT) and (distributed) Artificial centralized systems, taking into account functional and non-
Intelligence (AI). Despite several efforts towards standardizing
Edge Computing services and their ecosystem-driven interactions,
functional application requirements. At the same time, such
especially from the telecom world and open-source communities, mechanisms need to account for exploiting the advanced
automation in deployment, operation and interoperability of processing and Artificial Intelligence (AI) / Machine Learning
Edge Computing services is still in an immature state, making (ML) capabilities that smart devices, such as smart phones,
practical Edge/IoT scenarios that involve multiple application smart cameras and robots are nowadays equipped with, which
services and technologies, complex and cumbersome. This paper however present great variability in supported technologies and
presents our approach of an integrated platform aimed to simplify protocols. Although several efforts are ongoing for providing
the establishment, management, control and monitoring of edge edge-enabling mechanisms and for standardizing the archi-
computing services with a particular focus on the IoT domain. tectural frameworks and the ecosystem interactions of edge
Central in this effort are the ability to (I) make use of existing services (e.g., ETSI Multi-access Edge Computing (MEC) [3]
functions and modules in new edge computing services and (II)
to seamlessly integrate service function chain components with
from the Telecom sector and Linux Foundation Edge1 from
remote backend services and with locally available Edge/IoT open-source communities), it still remains a challenging and
devices for on-device processing. Characteristic examples of complex task to automate the deployment, integration and
desired functions and local processing tasks, include data stream operation of a set of diverse fit-for-purpose functions at the
analytics, event-driven workflows and Machine Learning tasks edge towards supporting end-to-end practical use cases for
with an emphasis on video stream analysis for Object Detection consumers and enterprise customers.
and/or tracking, for which we provide a deployment architecture
of our approach. To respond to the above-mentioned edge automation chal-
lenges, we introduce the TANDEM approach, presenting a
Keywords- Edge Computing, Function as a Service novel Edge platform architecture and supporting mechanisms
(Function-as-a-Service (FaaS)), Service Chain Orchestrator, for its user-centered and robust operation. The aim of TAN-
Artificial Intelligence (AI), Internet of Things (IoT) DEM is to allow end-users to develop, execute and manage
edge services without the cost and complexity of building and
I. I NTRODUCTION maintaining the infrastructure, while at the same time being
The rapid growth of connected Internet of Things (IoT) de- offered the capability to utilize the off-the-shelf, reusable and
vices and the corresponding dramatic increase in data volume, integrate-able services it provides. TANDEM supports new
which is expected to increase even more with the proliferation forms of computational models and focuses on an innovative
of connected devices in 5G networks, is causing latency and aspect of edge computing not yet adequately explored, and
bandwidth bottlenecks, when all data needs to be transmitted specifically the composition of services and applications from
to core nodes for centralized processing. As a response, structural elements that can operate on different edge nodes. It
Edge Computing [1] has gained traction as an alternative takes into account research approaches and standards for the
distributed paradigm, bringing data processing and storage design of a system which automates the execution of service
closer to end users and data sources, exploiting available chains and it proposes extensions for the creation of an inte-
compute and networking resources all along the cloud-to-edge grated edge computing system that seamlessly incorporates IoT
pathway and reducing this way response times and bandwidth, functions and AI / ML mechanisms. TANDEM covers business
as well as preserving privacy of data within their generation aspects and manages the transformation of a technical service
environment. Edge Computing is rather complementing than into a trade-able service (i.e., a product) by adding elements
replacing the already established paradigm of (centralized) such as pricing and Service Level Agreements (SLAs) and the
Cloud Computing and organizations that use them together service and product lifecycles.
will benefit from solution synergies that maximize the value
of both centralized and decentralized models [2]. 1 LF Edge: https://www.lfedge.org/
As a realization of our approach, the TANDEM platform such as Intel Smart Edge2 for performant and optimized
supports new serverless and Platform-as-a-Service (PaaS) edge service deployment on premise and at network edge, and
computing models. Specifically, TANDEM users can use its Edge Gallery3 with additional emphasis on fereration and
existing supporting and elementary services, as well as to interconection of third party applications. In this direction, a
combine existing services with their own custom services novel approach is also presented in [9], where it is suggested
and create complex service/function chains, using the plat- to abstract the northbound API of Edge Platforms into using a
form’s graphical tool. The target is typical edge computing PaaS edge service delivery model, for seamless interconnection
environments, such as Industry 4.0, connected vehicles and with 5G Orchestrators.
smart domains (Smart Home, Smart City, Smart Grid, Smart
Farming etc.) with an emphasis on IoT aspects [4], where III. T HE TANDEM A PPROACH
devices’ interconnection and management, as well as multi-
protocol and multi-data type support are common challenges. TANDEM focuses on an innovative aspect of edge com-
In such environments, TANDEM fosters the seamless integra- puting that has not yet been adequately explored, i.e., the
tion of data analytics services, using Analytical Processing composition of services and applications from structural ele-
and AI / ML technologies to predict future values (e.g., ments that may be available or must operate on different edge
of sensor measurements and system usage) and to provide nodes. TANDEM uses AI to predict future utilization of each
value-added edge services such as video stream analysis for node’s resources and to decide i) on the service instances that
Object Detection. Both reusable and custom services on the will be activated in each node and ii) on the edge service
TANDEM platform utilize available underlying edge resources lifecycle management, taking into account the accessibility to
and connected devices, as well as auto-scaling and auto-healing IoT devices from each node, the resource requirements of each
TANDEM mechanisms, and end-users can easily develop Web service, the agreement details with the customer (SLA), the
or mobile applications by calling the Application Programming participation of a service in a service chain and service affinity
Interfaces (APIs) of TANDEM platform services. and co-location requirements as well as the reduction as much
as as possible of energy consumption while satisfying the set
The remainder of the paper is structured as follows: In requirements.
Chapter II, the related work is presented and the differentiation
of TANDEM is explained. Chapter III provides an overview A. PaaS, FaaS and Backend-as-a-Service (BaaS) in TANDEM
of the TANDEM approach, the benefits it brings and the
technological concepts that it integrates at the edge. Chapter IV FaaS has received significant traction during past years, as
describes TANDEM system architecture that achieves drastic a service-based model that alleviates the need for application
simplification of the services development, management, mon- developers to worry about underlying infrastructure, auto-
itoring and extension processes in virtualized edge infrastruc- scaling/healing, and various support services for data persis-
tures. Chapter V describes a specific AI (Object Detection) use tence, authentication, event queuing etc., since all these are
case, which TANDEM offers as a service and presents a series handled by cloud providers. Thus, application developers are
of tools that TANDEM utilizes. Finally, Chapter VI concludes only left with the task of embedding their application business
this paper. logic into reusable fragments of code, i.e., the functions,
which are usually orchestrated into event-driven workflow
compositions.

II. R ELATED W ORK We adopt a similar approach for TANDEM custom services
and we expose a usable FaaS API and corresponding GUI for
In the context of (Big) Data Analytics automation platforms end users (i.e., customers) to define their own functions, or
for cloud, fog and IoT environments, several relevant EU to select among available ones pre-onboarded to the system,
research projects have been studied. PrEstoCloud [5] will make implementing the edge service logic that corresponds to their
research contributions in the cloud computing and real-time application requirements. These functions can interact with
data intensive applications domains, in order to provide a dy- each other, as well as with the readily available TANDEM
namic, distributed, self-adaptive and proactively configurable support services and TANDEM platform handles all the regis-
architecture for processing Big Data streams. RADON [6] tration and discovery of such services, the data storage, event
aims at creating a DevOps framework to create and manage queuing and dispatching services as well as the composition
microservices-based applications that can optimally exploit and interlinking automation of such services and their opti-
serverless computing technologies. RADON applications will mised deployment on available infrastructure.
include fine-grained and independently deployable microser- To achieve the above-mentioned platform automation fea-
vices that can exploit Function-as-a-Service (FaaS) and con- tures, TANDEM handles the exposure and availability of FaaS
tainer technologies. RAINBOW [7] will design and develop Platform support services while introducing an important nov-
an open and trusted fog computing platform that facilitates elty on their automated deployment. More precisely, TANDEM
the deployment and management of scalable, heterogeneous advocates a PaaS-based service delivery model, whereby the
and secure IoT services and cross-cloud applications (i.e., enablement and enrichment of available Edge nodes with
microservices). Finally, ELEGANT [8] aims to develop a novel specific FaaS support services, is managed in a declarative
software solution that addresses key challenges facing IoT and fashion that abstracts their deployment and operation details.
Big Data: interoperability, reliability, safety and security.
2 https://www.intel.com/content/www/us/en/edge-computing/smart-
When it comes to existing Edge / MEC Platform solutions, edge.html
there are noteable opensource projects initiated by big vendors, 3 https://www.edgegallery.org/en/pc-edgegallery-overview/
The optimization, resilient and secure operation of support TABLE I: TANDEM Entities and Roles
services are also undertaken by the TANDEM platform, which
respects underlying resources’ characteristics and state and Entity Description
performs all necessary steps for configuring the FaaS-enabled Infrastructure Provides the infrastructure (network, compute and storage), for
Provider TANDEM platform operation. Could be a Telecom Operator
edge node—from virtualization infrastructure preparation, to with computing and/or storage resources next to Base Stations
platform services’ readiness and to (FaaS) application-layer Service Installs the TANDEM system on its own infrastructure or
Provider that of an Infrastructure Provider and offers it as a service.
services’ orchestrated activation and interoperability. It makes adjustments or adds functions according to its own
needs (internal use) or its customers’ needs
Regarding the interoperability of services running on egde System Sup- Installs TANDEM on behalf of customers and supports them
nodes, with centralized (usually legacy) services including port Service daily in its operation, adjustments or additional functions and
Provider services.
authentication and authorization, master data management, Software The TANDEM system prototype will be implemented through
user and content management, we adopt a BaaS approach. Provider open source software and will be given free access through
According to this approach, interfacing from edge nodes free access repositories, e.g. GitHub.
Customer An entity (individual, company) registered on the TANDEM
towards backend services is automatically configured and platform and using its services. i.e., installs their own custom
policed in a secure fashion, using Quality of Service (QoS)- services that can cooperate with TANDEM services or creates
aware tunneling mechanisms and whenever applicable and complex services that will combine functions of existing ones

feasible, trade-off analysis (e.g., latency versus bandwidth) is


continuously conducted for deciding whether to launch local
copies of backend services on the edge node.

B. Relation to standards
In relation to standards and architectures, TANDEM is
directly harmonized with most standards, such as ETSI Net-
work Function Virtualization (NFV) Management and Orches-
tration (MANO), in terms of orchestration of the lifecycle
management of PaaS-driven support services as Virtualized
Network Functions (VNFs). In addition, we follow ETSI MEC
principles in terms of the edge computing architecture, as well
as the information model for describing edge services, using Fig. 1: AI and Artificial Vision from the core cloud to Edge
the MEC Application Descriptors (AppD) for the description clouds
of service attributes. and the TM Forum .
IV. A RCHITECTURE OVERVIEW
Moreover, the TANDEM service model extends the ETSI
MEC model with elements from corresponding models of the The TANDEM architecture takes into account the Euro-
TM Forum4 in terms of the description of the services, the pean Telecommunications Standards Institute (ETSI) NFV [13]
products created by the services, the product offer, and the reference architecture as well as the ETSI Multi-access Edge
product order. Finally, the service model is also extended with Computing (MEC) [3] architecture, while adding modules that
elements from the Open Service Catalogue Manager [10], support all necessary functions to fulfill the above usage sce-
aiming at additional coverage of business aspects, i.e., how narios. The key new feature TANDEM introduces, is the ability
a service is offered to a customer. Thus, TANDEM manages to create more complex services and applications by combining
the transformation of a technical service into a trade-able one services and integrating devices from different edge hosts or in
(product) by adding elements such as pricing and SLAs, and the cloud. The architecture of the TANDEM platform is based
also manages the service, product supply and product order on microservices [14]. Its edge nodes (hosts) are organized
lifecycles. in one or more edge clouds, each of which forms a cluster.
Figure 2 shows the levels of TANDEM architecture. It consists
of three basic levels, the System Level, the Edge Computing
C. TANDEM Entities and Roles Level and the Device Level. An additional intermediate level
between system and edge can be the Edge Cloud Level which
The basic entities and roles of TANDEM are described undertakes the management of an edge cloud.
in Table 1. TANDEM Service Provider installs the TANDEM
At the System Level reside the TANDEM functions that
(open-source) system on its own infrastructure or that of an
coordinate resources (computing, storage and networking),
Infrastructure Provider and has revenues from its provision as
devices and services of all nodes at the edge. The user can
a service and from new services requested by Customers. Cus-
request the use of services, post their own services as well as
tomers can address a TANDEM service provider or have their
use a combination of third-party services. This level includes
own TANDEM system installation. In this case a TANDEM
also user management, pricing policies and billing functions.
System Support Service Provider could install and customize
TANDEM and provide daily support. The Edge Cloud Level, when available, includes the TAN-
DEM Edge Platform Manager with functions for application
4 https://www.tmforum.org/resources/interface/tmf623-sla-management-api- life-cycle management as well as monitoring and resource
rest-specification-r14-5-0/ management and forecasting future resource needs. It contains
the Virtual Infrastructure Manager (VIM), that undertakes The role of the Edge Orchestrator in an NFV architecture
the integrated management of the virtualized infrastructure based on [13] could be played by the NFV Orchestrator
(e.g. installation and configuration of virtual machines and (NFVO) component of the NFV Management and Orchestra-
containers). When this level is not available, its functions are tion (MANO).
supported by the other 2 levels.
2) Service Orchestrator: It manages all the functions re-
At the Edge Computing/Node Level we can distinguish lated to the services provided to the system clients. Cus-
the TANDEM Edge Platform, which includes all the services tomers have access to it through the TANDEM Service Portal.
offered by TANDEM to customers, IoT support services (e.g. Through the Service Orchestrator customers can use a service,
device management services), as well as services that have obtain a list of services provided and their pricing and get in-
been installed by TANDEM customers. formation about the services they use and their current billing.
The Device Level includes all the devices connected to The System Administrator uses the Service Orchestrator to
TANDEM (sensors, cameras, etc.). These devices may have centrally manage TANDEM services and related resources.
processing capabilities, in which case they could host TAN- The basic functions of the Service Orchestrator include
DEM microservices (e.g. robots). TANDEM can receive data 1) managing the services located on edge hosts and their
from these devices (e.g. measurements from sensors) and can billing information, 2) connected devices, user and privileges
in turn send commands or deploy microservices to them. management, along with pricing models, 3) create service
chains by combining multiple services and 4) selecting edge
hosts for application deployment.
The Service Orchestrator consists of the following mod-
ules:
The TANDEM Service Catalogue is responsible for
registering and storing the services. The information stored
for each service includes resource requirements, its type and
characteristics (functional and quality) and how it is made
available for consumption. The basic functions are a) Service
registration, b) Service discovery and c) Service management.
The Service Chain Orchestrator(SCO) searches the Ser-
vices Catalogue to find the services needed to implement a
Fig. 2: The basic levels of TANDEM architecture. specific application. Then, according to the way the services
are defined and their APIs, it is able to perform a composition
Figure 3 shows the TANDEM architecture based on the of services (service chains). It undertakes the orchestration and
general ETSI MEC architecture, along with its components connection of the selected services in order to create a single
which are explained below. application.Connected services can be deployed at different
nodes at the network edge. Through the graphical interface, the
A. System Level users can see the devices and services available and their APIs,
configure the devices and services, and finally define how the
Functions at this level are used by TANDEM’s System services are connected to each other and to the devices.
Administrator in order to manage and monitor the use of
resources and devices and the operation of applications on The Service Manager is responsible for the management
all TANDEM nodes. Also, by the Customers and Service of services and their life cycle centrally for all TANDEM
Engineers, previously described in Table 1. nodes. TANDEM uses AI to predict the future use of resources
at each node and to decide on the instances of the services
This level hosts the TANDEM System Manager that acts that will be activated in each of them and what their life cycle
as the coordinator of the functions performed at the TANDEM will be, taking into account the accessibility of IoT devices
level. Specifically, the Manager is responsible for the proper from each node, the service resource requirements and the
operation of TANDEM, as it functions as the orchestrator available resources, the agreement with the customer (SLA)
node for all the TANDEM edge nodes. This level contains and the participation of a service in a service chain.
the TANDEM Service Portal which is the only entry point to
third parties in TANDEM functions offered by Edge Orches- The Central Device Manager is responsible for the overall
trator and Service Orchestrator. The Tandem System Manager device management for all TANDEM nodes. The administrator
constitutes of the two components below: has an overview of all devices, their features, their use and
status, and can manage their configuration. Lastly, the User
1) Edge Orchestrator: It manages the infrastructure, re- Management, Pricing and Billing Services. TANDEM has
sources and operation of the services and is used by the functions for (a) managing users, user roles and access rights,
TANDEM system administrator. Its primary functions include (b) defining and managing pricing policy, and (c) creating and
1) service, resource and application management (eg. life- presenting detailed billing information.
cycle, migration) at the edge, 2) uploading and registering
application packages on edge hosts and 3) maintaining an 3) Service Portal: It is the point to access TANDEM fea-
overall picture of the TANDEM system, with its available tures offered by Edge Orchestrator and Service Orchestrator.
resources and topology. Interested parties can use the portal to make their services
Fig. 3: The architectural components and core technological stack of TANDEM

available, create service chains or gather business and technical V. AI USE CASE AND DEPLOYMENT ARCHITECTURE
information about the available services.
One modern analytics scenario constituting a candidate for
edge computing is Object Detection. Computer vision and
video analysis are projects that usually require a large amount
B. TANDEM Edge Platform Level of hardware resources as well as vast volumes of data (images
and video streams) that need to be processed in real time,
At the top level of TANDEM Edge host are services that in most of the cases. The TANDEM platform offers such a
can be directly matched to end-user requirements, and can be service in the form of applying object detection algorithms to
the core services in end-user applications, and also be used as supervise a parking area. Live video streams from cameras are
part of a chain of services. These services, called Application used as an input to the TANDEM object detection service and
Services, are the basic services offered to the clients and they useful information can be obtained for the operation of the
are: 1) Data Analytics from IoT devices, 2) Object Detection parking such as detecting the number of free parking spots,
from a video stream, 3) Time Series Forecasting from IoT wrongly parked cars, accidents that might require to create
devices like sensors. alerts, provide assistance etc, in real time.
Following TANDEM’s approach of deploying object detec-
Furthermore, there are also the IoT support services re- tion to the edges, it’s possible to take advantage of resources
quired to create applications, for example for accessing data that might be available at the edges and also minimize require-
from external devices. These services are: 1) Device discovery ments for bandwidth as only the extracted information of the
and management services, 2) Scheduling Services, 3) Alerts services need to travel through the network, in comparison to
& Notifications services and 4) Rules Engine services. the whole video stream. Also, privacy issues are avoided since
the images of people are only handled locally.
These services can be based on services provided by an
IoT platform such as EdgeXFoundry [15]. A support service An end user may use the TANDEM service catalogue to
is the Message Queue used for the communication between find information about the object detection service as well
services that are part of a complex application. The TANDEM as the support services (for example an alerting service).
Edge Host includes also System Services that correspond to The services can be configured in regard to what output
the MEC Necessary Services like the Location service or the they will provide (for example the object detection service
Traffic Management service. There are also services that the can give updates about the number of objects of interest, or
End User has installed in their infrastructure (User Application provide a video stream with bounding boxes around regions
Services). These services can communicate with TANDEM of interest) and their interconnection (for example, the rules
services to use their functions or to offer functions. that cause an alert of a specific kind). Once the number of
services and the application requirements have been decided,
The TANDEM Edge Platform Manager has functions for the TANDEM service chain orchestrator can be used to create
monitoring and management of one or a cluster of edge the appropriate connections and deploy the services at the
hosts(edge cloud). It handles the service requirements and life proper edge nodes. The TANDEM service manager decides the
cycle and has AI mechanisms for resource usage prediction. optimal deployment location, taking into account the available
resources, the expected requirements and the quality of service through the selected usage scenarios. Particular emphasis will
requested by the end user. It is also responsible for the service be placed on the usability of the Service Chain Orchestrator
life cycle. Should one of the services fail, or should the (SCO) tool who undertakes the orchestration and connection
TANDEM Service Manager conclude that its quality of service of pre-existing services in order to create new, more complex
cannot be matched by the current installation, a new instance services and on the introduction of AI in the management of
will be created in an edge node that fulfills the requirements. services and resources.
Each time an object detection model is deployed for a new
parking application, the end user may decide to retrain the ACKNOWLEDGMENT
model with images from the new deployment site. By provid- This work is supported by the TANDEM (Function and
ing annotated images from their own parking lot, the end user Device as a Service Support for Edge Computing) project,
may request the model be retrained on the edge(for privacy co-financed by the European Union and Greek national
reasons) on the new dataset be added to the TANDEM model funds through the Operational Program Competitiveness, En-
repository to be available to other end users. Additionally, in trepreneurship and Innovation, under the call RESEARCH
the case of more than one deployments of object detection CREATE - INNOVATE (project code: T2E∆K-02825)
services in different sites, the services can also work towards
improving their robustness by implementing Machine Learning R EFERENCES
schemes such as federated learning.
[1] Ashkan Yousefpour, Caleb Fung, Tam Nguyen, Krishna Kadiyala,
Figure 3 summarizes the different tools and technologies Fatemeh Jalali, Amirreza Niakanlahiji, Jian Kong, Jason P. Jue, All
used by TANDEM platform in order to allow end users one needs to know about fog computing and related edge computing
to construct applications with functionality like the one de- paradigms: A complete survey, Journal of Systems Architecture, Volume
98, 2019, Pages 289-330
scribed above. At the lower level, devices communicate with
[2] Bob Gill, David Smith,”The Edge Completes the Cloud: A Gartner
TANDEM edge level either directly with REST or through Trend Insight Report”, 14 September 2018 ID: G00360847, Analyst(s):
a message queue (MQTT). A TANDEM edge host (edge Bob Gill, David Smith
level) is managed by Kubernetes [16] and can communicate [3] ETSI MEC, https://www.etsi.org/technologies/multi-access-edge-
with series DBs (eg Prometheus) and visualization tools (eg computing
Grafana) in order to store and expose the data from the [4] Sifakis, J., 2018. System Design in the Era of IoT—Meeting the
devices. Moreover, the edge host can utilize a number of Autonomy Challenge. arXiv preprint arXiv:1806.09846.
analytics tools (eg TensorFlow [17]) in order to use AI models [5] Proactive Cloud Resources Management at the Edge for
provided by TANDEM platform, partake in training them or Efficient Real-Time Big Data Processing (PrEstoCloud),
https://cordis.europa.eu/project/id/732339
even train new ones. At a higher level, the service orchestrator
utilizes workflow engines (eg Argo Workflows [18]) to create [6] Rational decomposition and orchestration for serverless computing
(RADON), https://radon-h2020.eu/overview/
proper relations between services according to a user’s desired
[7] An open, trusted Fog Computing Platform facilitation the deploy-
workflow and quality of service. ment, orchestration and management of scalable, heterogeneous and
secure IoT services and cross-cloud apps (RAINBOW), https://rainbow-
VI. C ONCLUSION h2020.eu/rainbow-platform/
[8] Secure and Seamless Edge-to-Cloud Analytics (ELEGANT),
This paper described in detail a new Edge Computing https://cordis.europa.eu/project/id/957286
platform architecture, which was designed to support practical [9] A. Lekidis, V. Theodorou, N. Psaromanolakis, C. Guerrero and D.
application scenarios that could benefit from the existence of R. Lopez, PiEdge: An Edge-Driven PaaS Model for Network Slicing
computational, storage and network resources closer to the Automation, 2021 Joint European Conference on Networks and Com-
data production and consumption points, covering the needs munications & 6G Summit (EuCNC/6G Summit), 2021, pp. 335-340
of different business sectors. [10] Open Service Catalogue Manager,
https://openservicecatalogmanager.org/
TANDEM is a platform which transfers computing work- [11] Qi Xia, Winson Ye, Zeyi Tao, Jindi Wu, Qun Li, “A Survey of Federated
load to the edge taking into account requirements such as Learning for Edge Computing: Research Problems and Solution”, To
delay, security, reliability and extensibility, while at the same appear in: High-Confidence Computing (2021)
time it optimally exploits the system’s capabilities. [12] Bosch IO whitepaper: Edge computing for IoT: A guide on
how edge computing complements the cloud in IoT, November
In conclusion, the TANDEM edge computing system was 2020, [Online]: https://bosch.io/wp-content/uploads/20201127 White-
designed in order to support the general usage scenarios Paper Edge-computing EN.pdf
listed in TABLE II. The expected benefits of TANDEM will [13] ETSI GS NFV-IFA 011 V2.1.1, “Network Functions Virtualisation
be demonstrated through the usage scenarios that will be (NFV); Management and Orchestration; VNF Packaging Specification”,
October, 2016
implemented and concern (a) the remote surveillance using
[14] The microservice architecture, https://microservices.io/
the Object Detection service, (b) the content caching at the
[15] The Preferred Edge IoT Plug and Play Ecosystem-Enabled Open
edge, so as to avoid requesting it from a central server, Software Platform, https://www.edgexfoundry.org/
and (c) the support of robots operation and cooperation in
[16] Production-Grade Container Orchestration, https://kubernetes.io/
Industrial Production in the framework of Industry 4.0. These
[17] An end-to-end open source machine learning platform,
use scenarios will be implemented by TANDEM in order to https://www.tensorflow.org/
achieve an evaluation and quantification of the benefits from [18] Argo Workflows - The workflow engine for Kubernetes,
a service perspective (e.g. response time reduction). https://argoproj.github.io/argo-workflows/
Future work includes the complete implementation of
TANDEM system elements and the system detailed evaluation

You might also like