SIESTA, SMEAGOL, VASP and BlueGene/L

Well, some of you have been regular readers (and commentors) for my journal in this part of the e-world and such of those have known that I have been in Europe – Jülich, Germany – to be precise. However, while chatting with few of my friends, I noticed an incoherent idea as to why yours-truly is in Germany any way… Some thought I was here on a pleasure trip/vacation, some others thought I had too much money and free time to burn, some more thought I was here for a photographic expedition, while only few have known the real reason…

We (as in our research group at Michigan Tech) have had a trial account on JUMP and JUBL in Jon von Neumann Institute of Computing, Jülich. But not having enough success, at least until recently, to get SIESTA/SMEAGOL compiled/running on our own beowulf linux cluster prevented me from pursuing similar attempts on JUMP/JUBL. Also, my inability to understand, let alone solve, the error messages spit out during many a futile attempts (on JUMP) didn’t really help much and neither did the 6 hour time difference between Houghton and Jülich.

Work week at the NIC/CBB got off to a most unlikely of starts, but a very pleasant one indeed – Champagne Party!! I don’t recall any meeting, let alone the first thing on a Monday morning, to begin with sipping Champagne. This party was in celebration of CBSB 07 success but like every such meeting, it also did a detailed post-mortem of the workshop itself – what went right, what went wrong, what was done better compared to previous edition, what could be done better in subsequent editions, possible locations for CBSB 08 … and so on.

Much of my working time at NIC/CBB was spent in making many parts of the source code XL Compilers compliant and fortunately enough, SMEAGOL compiled and test runs successfully finished on JUBL around 4:58pm Friday! Otherwise, weekend ahead would have been really loooong and so would have been the return journey. A successful compilation/execution of SMEAGOL implicitly implies the same about SIESTA and as an added bonus, one other group that successfully compiled and tested VASP on JUBL sent us the Makefile (and are willing to help us more, if need be) 🙂



SIESTA, SMEAGOL, VASP ... BlueGene/L

Our test calculation running on JUMP



SIESTA, SMEAGOL, VASP ... BlueGene/L

JUBL – 8 racks of IBM BlueGene/L



Working and discussing things with Jan as well as almost every member of the NIC/CBB group (discussions weren’t always about computers and/or sciences) has taught me many things: I think I know a bit more about IBM BlueGene; understand the term compiler strictness (I had very little or no such understanding when Dr. Roberto Orlando had mentioned it a in ASCS workshop); understand the compile-time error messages better and know how to fix some of them at least. Probably the most important lesson of all was the following:

Hope (that things will somehow work out for good) – when combined with patience, perseverance and practice – is a really good thing



PS: If you/your group have an access to BlueGene/L at Jülich or elsewhere and are trying to get these programs working, do not hesitate to contact for Makefile tips, source code editions, etc. You may just post your question as a comment in the form below and I will respond at my earliest.

9 Replies to “SIESTA, SMEAGOL, VASP and BlueGene/L”

  1. Hi,
    Have been trying to get vasp compiled on out BG/L system for a few weeks now – after resolving a few issues with library conflicts between the lapck routines that come with vasp and the IBM essl routines, managed to get it compiled BUT:
    At runtime it fails with a bunch of NANs which appear to be coming out of the summation routines in mpi.f.
    I would welcome any insight you have into getting it running – a makefile would be great, I assume you have used the xlf compilers? But any information you have can only help.

    Cheers,
    Nicola

  2. Hey, Gowtham – I’d like to hop on line to get that person’s contact info. Based on what I have been finding a well scaling version of VASP on the BG/P I have access to will be much easier than CPMD for the same problem.

    Thanks!

Comments are closed.