Tuning in the cluster environment
To make sure that programmes can fully utilise the features of modern CPUs, extensive tuning of the application is required. Numerous tools are available to support this process - special compilers that produce fully optimised code and profiling tools that can be used to analyse the runtime behaviour of the programmes. | Overview |  |  | Performance improvement |  | | Solutions | | | Benchmarks | |
Advantages of an overview
Optimising the programme code is the key problem faced by programmers. Weaknesses and bottlenecks need to be clearly identified so that appropriate intervention can be made to counteract them effectively. The same applies to localisation of errors in routines that cause programme aborts.
Powerful analysis tools are the answer to this challenge and transtec can offer a comprehensive range of these tools:
Improving performance
The compiler is the most important tool for programming custom software. The compiler must create a programme that is optimised for the relevant platform. However, the quality of the binary code differs significantly from one compiler to another. Because of our years of experience, we recommend commercial solutions >> from
- Intel
- Pathscale
Both manufacturers offer a comprehensive range of tools for compiling, improving performance and analysis. transtec can help you choose the right product for the architecture you are using.
Commercial solutions
Both Intel and Pathscale are leading suppliers of sophisticated compiler software. Intel compilers are among the most widely used commercial compilers and support the greatest variety of CPU types. Pathscale is rightly considered one of the stalwarts of the compiler business. Their key benefit is their optimised code for specific CPU types, e.g. AMD quad core CPUs, promising even greater performance.
transtec offers the following choice of compilers and tuning tools:
- Intel C/C++ and Fortran compiler
- Intel Math Kernel Library 10.0
- Intel Cluster Toolkit 3.1
- Pathscale Compiler Suite
A compiler should always be kept up to date to achieve optimum performance. This is done by purchasing update and support licenses. A key advantage of this is that you receive direct manufacturer support.
To supplement the compilers, Intel produces the Math Kernel Library and the Intel Cluster Toolkit. The library supplements the Intel compilers mentioned above with routines from four areas of scientific calculation.
- BLAS libraries with functions from linear algebra
- LAPACK routines
- ScaLAPACK (parallel LAPACK version)
- Fast Fourier transformation and random generators
The Intel Cluster Toolkit has been developed to meet special requirements of HPC clustering. It is compatible with both Windows and Linux-based HPC clusters. It includes:
- Math Kernel Library - cluster compatible version
- MPI Library
- MPI Benchmark
- Trace Analyzer/Collector for debugging MPI applications
On request, we will be pleased to advise you on the right choice of products and provide demo versions of the products where required.
transtec and benchmarks
There are two options for tuning the performance of a cluster - either tuning the hardware or alternatively the cluster software. Comparing the cost outlay for speeding up computing performance, it is normally worthwhile using professional software even for a small number of nodes. To be sure, you can start by using free demo versions. transtec can help you with the following services:
- Provision of demo software
- Consultancy on choice of appropriate compiler
- Creation of personal benchmark
- Consultancy on correct cluster configuration
