FY 2003 User Survey Results: Software

• Legend
• Satisfaction with Software (sorted by Average Score)
• Satisfaction with Software (sorted by Platform)
• Comments about Software:   28 responses

Legend:

Satisfaction with Software

Sorted by Satisfaction Rating

Satisfaction with Software

Sorted by Platform

Comments about Software:   28 responses

Debuggers and performance analysis tools:   8 responses

The debuggers are terrible. A colleague of mine described seaborg as "world-class hardware with dime-store software." and I agree. While the average software has improved tremendously (GNU tools, better editor support such as vim & pico, ...), the compilers and debuggers are a tremendous step down from the Cray. I have gone back to write statements in my code to get things working.

Debugging multiple CPU software on SEABORG is a rather painful experience -- the whole poe interface is nasty and the unless one is constantly using the debugger, the learning curve is non-trivial. The HELP pages on the Web page are fairly useful and perhaps could be improved even more.

Totalview is a painful debugger to use, but I don't know of any other good options for parallel debugging, so I can't complain too much.

Debugging tools are hard to use them.

totalview is basically unusable.

I have tried to use totalview for debugging OpenMP jobs. First it was a complete failure, it did not really work with OpenMP. Later I was told that totalview had been fixed and tried to use it again, and failed again. This time it worked with OpenMP but still could not work with allocated arrays in Fortran.

Several of the performance/debugging tools don't work for 64-bit applications. Totalview has ongoing problems with crashes and batch job debugging.

In my own experience, the performance analysis tools and totalview debugger have become more difficult to use and less informative for large complicated codes with many subroutines and Fortran90 modules.

IBM compilers and libraries:   7 responses

IBM's bugfixing is slow !

it can be hard to track down an error like a segmentation fault

The errors messages under MPI are not very informative. ...

The C++ linker gives thousands of incorrect "function already defined" messages. I have contacted NERSC about this and they have confirmed that it is a buggy linker and not my code. This makes linking become a nuisance. It would be nice if this could be fixed. Otherwise, I am very happy with the C++ compiler/linker.

Losing support from Kuck and Associates hurt my code's performance. The IBM compiler is not as fast. A C++ compiler that generates faster code would be useful.

There are very few C++ libraries for scientific computation available. ...

My main concern is about the speed of compilers. Although parallel make improves my compile time at O5 (which actually does increase the speed of my code by over 30%) my wait time is still almost 20 minutes. Also the compiler runs out of memory on compiling one routine since it cannot optimize a stencil (increasing memory has not resolved this problem). Would it be possible to provide a Power4 frontend for compilation?

Unix environment (shells, editors, GNU):   4 responses

I would like to have the following software supported: - gettext - libxml2 - popt - gsl

... I would like an alternative to Emacs for a full-featured terminal mode editor. Nano and pico do not have what it takes.

... I don't know how much priority this is given, but keeping AIX versions of public-domain software current is important. This is not merely emacs, and gzip, but the whole gamut of application development software, visualization software, and other things of the sort found on the UCLA AIX site. ...

I should mention, that I'm not a heavy user of NERSC, I'm basically setting up the program I wrote for others to run it and sometimes, when urgent results are needed, I may use the NERSC myself. Usually I'm running my programs at other facilities (e.g. LRZ in Munich). However, my just fresh impression from recent usage of the system (it's the first time I got to seaborg, prior I was using mcurie a few times) is that some software installations and the web pages are not the best they could be. Compiling etc., the new xlC 6, mpCC ... all that worked right away (I'm running the code on other AIX systems as well so it was ready for using the plattform) and I'm not using any complicated libraries. So that's not my subject here. But some of the goodies are in a bad state. Changing the shell for example. On the web pages "/bin/bash" is mentioned (and "/bin/ksh" as well --- by the way, on an IBM machine I would have expected to get a ksh by default -- as far as I can remember it was a csh here). But from my experiences I had to use /usr/bin/bash (due to /etc/shells or whatever). After that $PS1 isn't set (it's commented out in the default scripts ...). Get to my favorite editor "vim". It doesn't find it's runtime environment (take a look at the fallback of fallback for $VIMRUNTIME; after correcting that (for syntax color you immediately get into troubles concerning terminfo -- dump -Xany -Hv /usr/common/usg/bin/vim -> much better to have an ncurses around). There is no "less".

PDSF Software:   4 responses

No debugger working with STAR software

PDSF is rather thin on installed Perl modules, which means I have to maintain a lot of them in my home directory (and probably other folks are doing likewise), and leads to some serious spaghetti in my code.

Insufficient use for any useful comments. I am curious how you decide which "versions" of linux or say gcc to support and how you validate them - gcc 3.3 will soon be of interest to me.

We run RedHat 7.3 on our machines at LBL so it's easy to transfer programs and habits to the PDSF machines.

Satisfied:   4 responses

Since software resources are well documented on Nersc, it is the best place to try new softwares. I don't think drastic change is necessary.

... For the most part, however, I am happy with the software resources.

I used PETSc and TAO on several occasions and was very satisfied with both toolkits.

Our code has seamlessly ported to the IBM SP. I would like to use the debugging tools more, but have not had many bugs recently. I plan to do some minor code development and may invest more time to learn the debugger.

3rd party applications and libraries (viz, HDF):   3 responses

Great to have IDL 6.0 to test.

it would be good to have a possibility of interactive usage of typical chemistry visualisation software -molden, gopenmol, molekel, ecce

My only suggestion for improvement would be to keep users better informed on impending changes to the less common libraries (HDF comes to mind). Perhaps keeping the older versions around a little longer would be helpful also. ...

Other:   2 responses

Illustrative parallel code on the web site will be very useful

I don't use any other software except for Fortran, C, HDF, and FFTW.