Saltar al contingut Menu
Mapa
  • Inicio
  • Información
  • Contacto
  • Mapa

ARQUITECTURAS DE SUPERCOMPUTADOR (ASC)

Créditos Dept.
7.5 (6.0 ECTS) BSC

Profesores

Responsable:  (-)
Otros:(-)

Objectivos Generales

The goal of this course is to present the design of a supercomputer system, from the architecture of the individual core of a multicore chip in a shared-memory node, to the distributed memory system and the high bandwidth interconnection network. This knowledge will allow application developers to exploit the characteristics of the underlying system in order to write more efficient code.

Objectivos Específicos

Conocimientos

  1. To understand the basic concepts about microprocessor architecture.
  2. To understand the basic concepts about shared and distributed multiprocessor systems.

Habilidades

  1. To be able to exploit the characteristics of a target architecture to improve application performance.
  2. To identify potential bottlenecks in application performance.
  3. To identify the desirable system properties for a target application.

Competencias

  1. Be able to work individually or in group to discover the architectural issues that affect the performance of a specific application.

Contenidos

Horas estimadas de:

T P L Alt L Ext. Est O. Ext.
Teoria Problemas Laboratorio Otras actividades Laboratorio externo Estudio Otras horas fuera del horario fijado

1. Processor architecture.
T      P      L      Alt    L Ext. Est    O. Ext. Total 
8,0 0 0 0,5 0 8,0 0 16,5
1.1 Instruction set architecture: RISC, CISC, SIMD extensions.
1.2 Instruction pipelining, superscalar execution, out-of-order execution.
1.3 Cache memory hierarchy.

2. Multiprocessor architecture.
T      P      L      Alt    L Ext. Est    O. Ext. Total 
10,0 0 0 0,5 0 8,0 0 18,5
2.1 Multithreaded processors.
2.2 Shared memory multiprocessors.
- Memory coherence.
- Memory consistency.
2.3 Distributed memory multiprocessors.
- Distributed shared memory.

3. Interconnection networks.
T      P      L      Alt    L Ext. Est    O. Ext. Total 
6,0 0 0 0 0 6,0 0 12,0

4. Performance analysis tools.
T      P      L      Alt    L Ext. Est    O. Ext. Total 
8,0 0 6,0 0 0 4,0 0 18,0

5. Architecture conscious programming.
T      P      L      Alt    L Ext. Est    O. Ext. Total 
8,0 0 12,0 0 0 4,0 0 24,0
5.1 Improving single thread performance.
5.2 Improving multithreaded performance.
5.3 Improving multiprocessor performance.


Total por tipo T      P      L      Alt    L Ext. Est    O. Ext. Total 
40,0 0 18,0 1,0 0 30,0 0 89,0
Horas adicionales dedicadas a la evaluación 0
Total horas de trabajo para el estudiante 89,0

Metodología docente

Theory classes building up concepts in a structured fashion and setting out the commitment required for their practical application. The classes will give a perspective of the future trends. Laboratory classes focusing on co-operative work in order to consolidate concepts, skills and competencies.

Método de evaluación

Analysis of a set of selected architecture papers through a written report and an oral exposition. Optimization of a set of example applications for a candidate system architecture.

Bibliografía básica

  • J.P. Shen and M.H. Lipasti, Modern Processor Design: Fundamentals of Superscalar Processors, McGraw-Hill , 2005, First Edition.
  • J. L. Hennessy and D. A. Patterson, Computer Architecture: A Quantitative Approach, Morgan Kaufmann Publishing Co, .
  • Duato, J., Yalamanchili, S., and Ni, L., Interconnection Networks: an Engineering Approach, 1st. IEEE Computer Society Press, 1997.

Bibliografía complementaria

  • Proceedings of top computer architecture conferences: ISCA, MICRO, HPCA, ICS, Supercomputing, PACT, IPDPS, ICCD, ..., , .

Enlaces web

(Información no introducida)

Capacidades previas

-  Undergraduate courses in structured programming.
-  Knowledge of C.
-  Notions of MPI, OpenMP, Pthreads.
-  Notions of computer architecture.


Compartir

 
logo FIB © Facultad de Informática de Barcelona - Contacto - RSS
Esta web utiliza cookies propias para ofrecerle una mejor experiencia y servicio. Si continúa la navegación, entendemos que acepta nuestra política de cookies. Versión clássica Versión móvil