Publication: Estudio del rendimiento y la escalabilidad de aplicaciones MPI en entornos distribuidos utilizando SIMCAN
Loading...
Files
Official URL
Full text at PDC
Publication Date
2018-09
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citations
Exportar
Abstract
Este proyecto tiene como objetivo representar el comportamiento de aplicaciones MPI en entornos distribuidos simulados. En concreto, los entornos se representarán en la plataforma de simulación SIMCAN, una plataforma para modelar y simular entornos y aplicaciones distribuidas.
Las aplicaciones MPI son programas que trasmiten mensajes entre varios procesos las cuales pueden ser ejecutados en una o varias máquinas, consiguiendo ejecutar y procesar datos de forma paralela.
Para lograr el objetivo propuesto, las aplicaciones MPI se procesan en 3 Fases principales. La primera Fase consiste en el desarrollo de una biblioteca en lenguaje C con la que se genera un fichero de registro que contiene todas las llamadas MPI y de E/S ejecutadas por cada proceso. La segunda Fase consiste en integrar en SIMCAN el registro obtenido en la Fase anterior. Esto permitirá simular la aplicación MPI en distintos tipos de entornos distribuidos. Como resultado se genera un nuevo fichero de registro. La tercera Fase procesa uno o varios ficheros generados en las Fases anteriores y representa gráficamente la ejecución de la aplicación MPI. Con la representación obtenida es posible estudiar de forma detallada las ejecuciones de los programas MPI. Es importante destacar que así se puede comparar el comportamiento de la aplicación, ejecutada en entornos reales, con la ejecución de la misma aplicación en entornos simulados utilizando distintas configuraciones.
This project aims to represent the behaviour of MPI applications in distributed simulated environments. Specifically, the modeled environments will be represented in the SIMCAN simulation platform. MPI applications are programs that transmit messages among a number of processes, which can be executed in one or various machines. This enables the application to execute and process data in parallel. SIMCAN is a platform used for modelling and simulating distributed systems and applications. In order to achieve this objective, the MPI applications will be process in 3 main phases. The first phase, consists in developing a library in language C. This will generate a register file that contains all MPI and E/S calls executed by each process. The second phase consists in integrating the register file generated in the first phase into SIMCAN. This will enable the simulation of the MPI application in different types of distributed environments. As a result, this will generate a new register file. The third phase will process one or more generated files from the previous phases and will then represent graphically the execution of the MPI applications. The obtained representation makes it possible to study the executions of the MPI programs in detail. It is important to highlight that this enables the comparison of the behaviour of the executed application in real environments with the execution of the same application in simulated environments.
This project aims to represent the behaviour of MPI applications in distributed simulated environments. Specifically, the modeled environments will be represented in the SIMCAN simulation platform. MPI applications are programs that transmit messages among a number of processes, which can be executed in one or various machines. This enables the application to execute and process data in parallel. SIMCAN is a platform used for modelling and simulating distributed systems and applications. In order to achieve this objective, the MPI applications will be process in 3 main phases. The first phase, consists in developing a library in language C. This will generate a register file that contains all MPI and E/S calls executed by each process. The second phase consists in integrating the register file generated in the first phase into SIMCAN. This will enable the simulation of the MPI application in different types of distributed environments. As a result, this will generate a new register file. The third phase will process one or more generated files from the previous phases and will then represent graphically the execution of the MPI applications. The obtained representation makes it possible to study the executions of the MPI programs in detail. It is important to highlight that this enables the comparison of the behaviour of the executed application in real environments with the execution of the same application in simulated environments.
Description
Universidad Complutense, Facultad de Informática. Departamento de Sistemas Informáticos y Computación, curso 2017/2018