I have just switched to GEANT4.4.0 (from 4.3.2) and I experience a severe performance degradation. Programs that were able to treat 6x10^6 primary events per hour are now down to 3x10^5 and even 3x10^4 events per hour. I use exclusively low-energy electromagnetic processes and the degradation seems related to the cut (the lower the cut, the higher the degradation) How can there be such a difference between the two versions ? Richard
Dear Richard, as Makoto Asai pointed out in a previous reply to your enquiry, a patch correcting a problem found in Geant4 4.0 has been released, which speeds up the execution significantly. The problem was a general one, so I suggest you to verify whether with the released patch also your application improves its performance. I'd appreciate if you'd let us know whether any significant problem persists, even with the correction patch. Other possible sources of performance degradation in the last release have been investigated, and any further improvements, if achievable, would be available in next patches or release. As to lowenergy electromagnetic physics, we have in progress a major design iteration in the package, as well as physics extensions, that started last summer and will presumably continue for the whole 2002. Since this evolution is carried on through an iterative and incremental process, there may be occasional performance degradations (usually corresponding to the introduction of new functionalities), but we try to compensate them, whenever possible, with a following cycle devoted to performance optimization. We appreciate the patience of the package users, while this design iteration is in progress: we do our best to accomodate the needs of users requiring new or improved functionalities, as well as of those needing good performance, but, with the available resources in the group, sometimes we are obliged to split the extension and optimization cycles into two Geant4 public releases. Best wishes, Maria Grazia
Dear Richard, thanks for your feedback. We are investigating the performance of the Low Energy electromagnetic package specifically. By the way, two students of mine, working on a similar brachytherapy application, did not experience any major performance degradation with the latest Geant4 reference tag. We'll keep you up to date with any progress. Best wishes, Maria Grazia
The estimate to address this problem report takes approximetely four years' work to set up a testing and validation process in support of a performance improvement process. The Geant4 spokesman asked me to close this problem. Unfortunately, at the present time the closest response available in the problem report menu is "WONTFIX". In the meantime, however, many papers have been published based on Geant4 Low Energy Electromagnetic package, in spite of the poor performance mentioned in this problem report. We assume, therefore, that the poor performance does not prevent the usage of the package in a large number of use cases.
