Publication: Aceleración Hardware en entornos de Ciberseguridad
Loading...
Official URL
Full text at PDC
Publication Date
2019
Advisors (or tutors)
del Barrio García, Alberto Antonio
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citations
Exportar
Abstract
En este trabajo presentamos la construcción con Systems on a chip (SoCs) de un clúster de bajo consumo de energía y coste, que ejecuta un algoritmo de aprendizaje sobre el dataset introducido.
Actualmente existen diferentes tipos de clúster, en nuestro caso estará formado por SoCs de bajo consumo y coste. Las placas utilizadas son DE1-SOC de Altera, en este tipo de SoC se pueden introducir diferentes sistemas operativos. Para este proyecto hemos decidido usar la imagen que tiene la versión más reciente del kernel de Linux que es la 4.5.
Para nuestro trabajo vamos a ejecutar un algoritmo de aprendizaje automático, en concreto SVM sobre un dataset llamado KDDCUP`99 el cual contiene un análisis del tráfico en distintas redes en las que se han llevado ataques y otras en las que es un tráfico normal. Para esto nos ayudaremos de la librería libsvm la cual es una implementación en distintos lenguajes de una SVM, en nuestro caso en concreto usaremos la versión en C++ para realizar el entrenamiento de SVM en el procesador ARM de las placas DE1-SOC.
Finalmente, hemos procedido a hacer una comparativa de los datos de tiempo empleado en la ejecución y tasa de acierto que hemos logrado desde la utilización de una Workstation de altas prestaciones hasta la utilización del clúster con 6 nodos, donde hemos podido comprobar que la capacidad de acierto es mayor en la ejecución de la Workstation, pero sin embargo la ganancia de tiempo que logramos con el clúster de 4 ó 5 nodos hace que sean mejores opciones.
In this work we present the construction with Systems on a chip (SoCs) of a cluster of low energy consumption and cost, which executes a learning algorithm on the introduced dataset. Currently there are different types of clusters, in our case it will be formed by SoCs of low consumption and cost. The boards used are Altera DE1-SOC, in this type of SoC different operating systems can be introduced. For this project we have decided to use the image that has the most recent version of the Linux kernel that is 4.5. For our work we are going to execute an automatic learning algorithm, specifically SVM on a dataset called KDDCUP`99 which contains an analysis of the traffic in different networks in which attacks have been carried out and others in which it is a normal traffic. For this we will use the libsvm library which is an implementation in different languages of an SVM, in our case in particular we will use the C ++ version to perform SVM training in the ARM processor of the DE1-SOC boards. Finally, we have proceeded to make a comparison of the time data used in the execution and success rate that we have achieved from the use of a high performance Workstation to the use of the cluster with 6 nodes, where we have been able to verify that the capacity of success is greater in the execution of the Workstation, but nevertheless the gain of time that we achieved with the cluster of 4 or 5 nodes makes them better options.
In this work we present the construction with Systems on a chip (SoCs) of a cluster of low energy consumption and cost, which executes a learning algorithm on the introduced dataset. Currently there are different types of clusters, in our case it will be formed by SoCs of low consumption and cost. The boards used are Altera DE1-SOC, in this type of SoC different operating systems can be introduced. For this project we have decided to use the image that has the most recent version of the Linux kernel that is 4.5. For our work we are going to execute an automatic learning algorithm, specifically SVM on a dataset called KDDCUP`99 which contains an analysis of the traffic in different networks in which attacks have been carried out and others in which it is a normal traffic. For this we will use the libsvm library which is an implementation in different languages of an SVM, in our case in particular we will use the C ++ version to perform SVM training in the ARM processor of the DE1-SOC boards. Finally, we have proceeded to make a comparison of the time data used in the execution and success rate that we have achieved from the use of a high performance Workstation to the use of the cluster with 6 nodes, where we have been able to verify that the capacity of success is greater in the execution of the Workstation, but nevertheless the gain of time that we achieved with the cluster of 4 or 5 nodes makes them better options.
Description
Trabajo Fin de Grado en Ingeniería Informática, Facultad de Informática UCM, Departamento de Arquitectura de Computadores y Automática, Curso 2018/2019