LHC Sixtrack News
Status/Plans, 7th August 2012
First, many thanks for your continued support. From my/CERN side all has been running rather well and I am submerged by results. I now need to take some time to analyse them. In particular to decide between the two methods of computing the beam - beam effect. Then I shall probably submit several studies to do an intensity scan where I study the beam - beam effect depending on the size, and hence charge, of the accelerated bunch of particles. At the same time, I must finish the documentation of the "user" infrastructure so that my colleagues may easily use BOINC as they return from vacation. In addition I want to set up a dedicated "user" system "lxtrack" in order to provide disk space here and to try and keep up with the results as they are returned. I have to look at the Deadline problem for 10**7 turn jobs. I set a bound of 30 days for any WU....need to discuss with Igor is that is NOT what you see at home. Of course we really want a low bound to get results back quickly, but I also want to use older slower systems. We shall have to work out some sort of compromise. My attempt as 10**7 turns was probably a bit over the top, but I was keen to try it. We hope/expect to produce a valid MAC executable this week. I also need to add some new "physics", new elements, to Sixtrack as provided by a colleague. (Also need to add modifications for "Collimation" but they are not relevant to BOINC.) The next version should also support SSE4.1. I was very pleasantly surprised to win an NVIDIA TESLA C2075. The catch is that I have to use it and program it with OpenACC. There will doubtless be some hiccups intsalling the board and the necessary (PGI) software. I shall in fact try my "Tomography" application which already runs in parallel using HPF or openMP. If that works I shall seriously consider a multi-threaded Sixtrack (using GPUs or not) by tracking many more particles in each Work Unit. Non-trivial but rather exciting. I am just at the ideas stage here, but.....it would of course use multiple threads on a multiple core PC as well. A dream? Finally, I have to take time to publish my work on floating-point portability and reproducibility. I believe I might be the only person who gets identical bit for bit 0 ULP different results after many Gigaflops with 5 different Fortran compilers at different levels of optimisation.