Cypress CSC-1200T Guia do Utilizador descarregar pdf (Página 9)

Chapter 1. Introduction 9

1.3 Overview of the system

The Cray T3E system at CSC has currently 224 RISC processors for paral-

lel applications. In addition, there are 16 processors for system services

and for interactive use.

The T3E has a good user and programming environment. The sys-

tem feels like any Unix computer. You log in to the Internet address

t3e.csc.fi and end up in an interactive processor.

All processors share a common ﬁle system. The ps and top commands

can be used to look at processes on any processor (see page 29 for

details). The parallel nature of the machine is only manifested when

running parallel codes.

The single-processor performance is a critical factor in the performance

of parallel user codes. As on most RISC processors, user codes may

typically obtain only 5–15% of the maximum performance on each pro-

cessor. This is also true for the Cray T3E. However, in linear algebra

operations, the BLAS kernels can reach over 65% and LAPACK kernels

over 45% of the peak performance. See page 33 for more details.

The T3E has a remarkably fast communication network which makes the

machine a well-balanced system. It is quite easy to write parallel codes

that scale up to a hundred processors.

The Cray T3E is an IEEE-conformant system with 64-bit integer and

ﬂoating point representation by default. To conserve memory, you can

switch to 32-bit representation of integers or integers and ﬂoating point

values together.

The Cray T3E hardware is described in Chapter 3, and code optimization

is discussed in Chapters 5 and 6 (Fortran 90 and C, respectively).

The Cray T3E series is a product of Cray Research, which is a subsidiary

of Silicon Graphics, Inc.

1.4 Programming environment

The Cray T3E system oﬀers a versatile programming environment for the

users. There is a high-quality Fortran 90 compiler which can, of course,

be used to compile standard-conforming FORTRAN 77 programs as well.

Also C and C++ compilers are available.

Parallelization can be done using the Cray implementation of MPI (Mes-

sage Passing Interface), which has been optimized for the system. Also

the PVM libraries (Parallel Virtual Machine) are available. MPI is dis-

cussed in Section 7.2 (page 63) and PVM in Section 7.3 (page 68).

1 2 3 4 5 6 7 8 9 10 11 12 13 14 ... 123 124

Sem comentários

Cypress CSC-1200T Guia do Utilizador Página 9