<html>Dear all,<br />I'am doing phonons of supercells that take LLONNNGG time (more than 1000 modes).<br />For this I need a lot of restarts.<br />However, each restart seems to be longer and longer with the increasing number representations already done.<br />For example,<br />it takes 5h on 72 cores to begin to calculate the 'to be done' representations when 950 representations are done,<br />whereas it takes only 30 minutes when 100 representations are done.<br />Which means that for the policy of supercomputers that limit jobs to 24h, I loose 25% of cpu time.<br /><br />I suppose that qe reads all the ones that have be already done during the restart to check that they are all finished.<br /><br />I was wondering if there exist any trick to avoid this reading that is useless as the new representations are independant of the previous ones.<br /><br />Regards,<br /><br />Antoine Jay<br />LAAS CNRS<br />Toulouse, France</html>