A rich variety of software to extract the best performance from the SR11000 model J1
The latest UNIX OS with 64-bit addressing support and Hitachi FORTRAN and numerical calculation libraries optimized to extract the best hardware performance from the latest POWER5 architecture are provided. Many ISV packages are also provided.
|
|
General description of each product |
Basic control
 |
With 64-bit addressing, large scale calculations with 100 TB files and more than 100 GB memory, can be processed at high speed. |
|
In order to execute jobs efficiently, various job management functions are provided. The dynamic load distribution function dynamically assigns jobs to under-loaded systems and executes them. A function to designate job execution time and method can be used as a flexible system administration function, such as batch job execution at night.
Using these functions allows hardware resources to be effectively utilized. |
|
Job execution control functions such as job start/cancel/condition display, checkpoint/restart and accounting logs are also provided. |
|
Operating status monitoring functions, graphical display functions, tools for diagnosing processing performance problems, tools which analyze system resources and operating status and so forth are provided. Processing performance of the whole system can be optimized using these functions. |
|
Center administration functions such as budget and result management, accounting, and automatic system operation are inherited from the predecessor, the SR8000. |
Program development assistance
Message passing libraries such as MPI and MPL which are used for inter-task communication of parallel programs, parallel debugger, and execution status monitoring / analysis tools for parallel applications are provided as effective development and execution environments for C, C++ and FORTRAN parallel programs.
Parallel programs can be easily and effectively developed and job execution status can be monitored visually using these software tools.
Compilers (Language)
Provided languages include a C compiler conforming to industry standards such as ANSI C and OpenMP specifications, a C++ compiler conforming to ANSI C++, a FORTRAN compiler conforming to Fortran 77/90/95 standards and program development environments such as debuggers.
- Optimizing FORTRAN90
Provided as development/execution environments for making applications are an automatic parallelizing function using high level analysis on DO loops, and a program tuning assistance function which assists users with optimized coding at compile time, for example a diagnostics messages output function. In addition, this compiler is optimized so that the highest degree of hardware performance can be extracted from the SR11000 model J1.
- High level automatic parallelization function
High SMP system performance can be achieved without inserting directives using automatic parallelization based on high level DO loop analysis techniques.
- High-speed parallel processing function
High parallelizing efficiency can be achieved for an SMP system using the low overhead parallel processing function.
- Optimization diagnostic messages assisting program tuning
The parts of a program that behave as bottlenecks can be easily detected by outputting optimization diagnostic messages during compile.
- Other functions and features
|
Support for industry standard OpenMP 2.0 |
|
Optimization by utilizing the dynamic information derived from profiling |
|
Performance enhancement by extended automatic inline optimization |
Numerical calculation libraries
Numerical calculation libraries are provided that correspond to scientific and engineering applications that have achieved high performance in various fields.
- MATRIX/MPP
The matrix calculation subprogram library MATRIX/MPP is a subroutine library that fully optimizes hardware performance in functions frequently used in scientific and engineering, for example, simultaneous linear equations, eigenvalues, fast Fourier transforms and pseudo-random numbers. It can be used from programs in FORTRAN or C and not only does it have interfaces corresponding to shared memory type parallel processing, but it also has interfaces corresponding to distributed memory type parallel processing. Major functions are shown below
|
Basic array operations |
|
Direct methods for simultaneous linear equations |
|
Iterative methods for simultaneous linear equations |
|
Inverse matrix |
|
Eigenvalues and eigenvectors |
|
Fast Fourier Transform |
|
Pseudo-random numbers |
- MATRIX/MPP/SSS
The matrix calculation subprogram library - sparse matrix solution MATRIX/MPP/SSS is a library for large scale sparse matrices used in structural analysis. The library has interfaces corresponding to shared memory type parallel processing, and also interfaces corresponding to distributed memory type parallel processing. Major functions are shown as follows.
|
Skyline method |
|
Sparse solver |
|
Ordering function |
|
Iterative methods for simultaneous linear equatious |
|
Eigenvalues and eigenvectors |
- MSL2
The mathematical subprogram library MSL2 is a subroutine library that supports calculations that are frequently used in numerical calculation fields, for example, matrix calculation, function calculation and statistics calculation. It can be used from programs in FORTRAN and C. Since it has gathered high reliability methods, subroutine programs of several methods are provided and thus it has high adaptability to problems.
Trademark legend)
• UNIX is a registered trademark in the United States and other countries, licensed exclusively through The Open Group.
• POWER5 is a trademark of the International Business Machines Corp. in the United Statesand other countries.
• MPI is an interface convention set by the MPI Forum in the U.S.
• Other product and company names mentioned in this document may be the trademarks of their respective owners.
Hitachi Global Site 
