HPCx homepage
Services User support Projects Research About us Sitemap Search  
      line      
Atomic and Molecular Physics Biochemistry Chemistry Computational Engineering Numerical Algorithms Environmental Sciences
Computational Material Science New Application Areas Publications Euro HPC HPC Research Newsletter
               
home > research > hpc > technical_reports > HPCxTR0611
spacer
hr
spacer

HPC Research

spacer

HPCx Tools Survey

The purpose of the HPCx tools survey was to discover what software development tools users employ for their computational science research on a range of different systems: on their laptop or desktop; on medium scale parallel machines provided at departmental or university level; on national supercomputers such as HPCx. The aim was to find out what tools are actually used on HPCx, what tools it would be useful to install on HPCx, and what gaps there are in the availability of tools on high-end supercomputers.

The pool of survey participants was made up of twenty-one registered users of the HPCx system. The majority of these users are also developers of scientific software (mainly Fortran with MPI). The usage of code development tools across desktops, local and national HPC provision is relatively uniform: where version control is used, CVS is the preferred tool; the main development environments used are Emacs and other text editors; the preferred build management software is make; the main debugging tool is dbx/gbd (as well as some use of totalview on HPCx).

The most widely used numerical libraries are LAPACK and BLAS, as well as FFTW; NetCFD is considered to be the most important IO library. Less than half the survey participants use tools for parallel analysis (MPItrace and Vampir) and memory debugging (Valgrind, totalview and hmd); the complexity of the tools and the quality of the results are stated as the main reasons for the limited usage. Some users comment that MPI correctness checking tools, OpenMP job analysis tools and and effective memory debugging tools are missing on the national HPC system.

Half of the survey participants believe that lab or lecture based tool training is important or very important, whereas 28% think it is unimportant. The main concerns raised in the comments are the (often) varying levels of experience of course attendees, as well as the problem that courses are rarely available when a user needs to actually use the tool, and that therefore a good manual might be more valuable than a course. However, it is also pointed out that training can be advantageous for complex tools where manuals become unmanageable. The majority of users find that having access to the same tool suites on different levels of computing hardware is important. One comment however suggests that being forced to use different tools can have a positive effect, because different tools are more likely to pick up on errors.

Registered users of HPCx can find the survey and full details of the results here (scroll to the bottom and click on "Survey results: View").

spacer
hr
spacer
http://www.hpcx.ac.uk/research/hpc/technical_reports/HPCxTR0611.html contact email - www@hpcx.ac.uk © UoE HPCX Ltd