Novel heterogeneous deployment of distributed applications across the compute continuum with COMPSs

Industry 4.0 has caused a revolution in how data is created and analyzed. Great amounts of information are retrieved and generated non-stop by a myriad of sensors and small devices, which form the Internet of Things. For analytic workloads to keep up with the incoming stream of data, applications ne...

Descripción completa

Detalles Bibliográficos
Autor: Perez Mendizabal, Unai
Tipo de recurso: tesis de maestría
Fecha de publicación:2020
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2117/336188
Acceso en línea:https://hdl.handle.net/2117/336188
Access Level:acceso abierto
Palabra clave:Cloud computing
Computational grids (Computer systems)
compute continuum
cloud computing
edge computing
fog computing
distributed application
distributed framework
task-based framework
containerization
docker
container orchestration
compss
Computació en núvol
Computació distribuïda
Àrees temàtiques de la UPC::Informàtica::Arquitectura de computadors
Descripción
Sumario:Industry 4.0 has caused a revolution in how data is created and analyzed. Great amounts of information are retrieved and generated non-stop by a myriad of sensors and small devices, which form the Internet of Things. For analytic workloads to keep up with the incoming stream of data, applications need to run in a distributed manner, and, as such, many frameworks for distributed applications have flourished in the recent years. These frameworks normally require the use of specific APIs that harm and prevent the code usability and portability. Additionally, those are usually aimed at cloud environments and demand powerful resources to run. The edge and fog computing paradigms have proven to be useful due to their closeness to the data, which can provide real-time guarantees that the cloud is not able to. None of these distributed application frameworks aim at providing a transparent deployment across the compute continuum, thus losing the benefits of a truly distributed workload. COMP Superscalar (COMPSs) is a distributed application framework that uses a very simple API that allows the execution of sequential programs in a distributed manner. Although one of the key benefits of COMPSs is its deployment transparency, it lacks the features necessary for a heterogeneous deployment across the compute continuum. In this thesis, new features for COMPSs are introduced, which extend its transparent deployment and code portability capabilities to edge and fog through containerization, by means of Docker and Linux Containers (LXC). The setup process is reduced to the possible minimum thanks to the new lightweight images. A new integration of cloud platforms has also been designed and implemented to support the most modern, container-oriented cloud platforms such as Docker Swarm or Kubernetes. This has allowed to transparently run distributed applications in heterogeneous continuum environments, facilitating the setup while not introducing any kind of latency to the execution.