Dear Florian, concerning BenchExec, that's good news. In any case we need reliable information about the allowed memory size for the competition. Otherwise, memory settings like Xmx$((${STAREXEC_MAX_MEM}-40))G fail spectacularly: Today's test on new.q reports "Could not reserve enough space for 137397010432KB object heap". Best Dieter On Fri, 2024-06-14 at 08:49 +0200, Florian Frohn wrote:
I got new from the StarExec team. In the best case, BenchExec will be back at the beginning of next week.
If they don't manage to get it running, we'll ditch the memory limit completely (by setting it to an unrealistically huge value), and we'll only run one benchmark per node at a time. If a solver exhausts the memory of its node, it will start swapping, slow down, and time out, eventually.
I think both options are acceptable for us, even though the latter will double the time that's needed to run the competition, as we can only run one instead of two benchmarks per node. So in this case, we'll have to think about countermeasures (e.g., a smaller timeout).
-- Dieter Hofbauer dieter_hofbauer@web.de Mühlengasse 16, D-34125 Kassel, Germany phone (+49) 561 739 29 54