Good afternoon Paolo, > if you really use 2*250Mb out of 4Gb, you will not swap, unless your > machine is horribly loaded. If you are swapping, you are using more > memory than you think. How did you get that number? it is not easy > to know how much memory a process is really using.
I got it with top. "ps aux | grep ph.x" would also report that size. What really worries me is that, although the only processes using over 100Mb are mine (top again), and they are the only active ones, the kernel swaps memory instead of dumping some old memory. The cache page was about 300Mb. > Check with command "size ph.x" how big is the so-called "bss". I will follow your suggestion. > you mean g95, or really gfortran? these are two distinct projects Yes, g95, sorry. >>3) This is a bit off topic. Is it OK to use -npools while using >>processors in the same node or is it dumb? > > it is ok, but it won't save any memory: each process will allocate > the entire memory If there were enough memory... would that perform better than not using -npools? What I understood from the guide is that espresso will parallelise in kpoints according to npools and on wavefunctions on nprocessors/npools. It also seemed to imply that, as long as -npools divides the total number of kpoints, the performance increase should be linear. But maybe I am totally wrong. Thanks again for your time, Miguel -- ---------------------------------------- Miguel Mart?nez Canales Dto. F?sica de la Materia Condensada UPV/EHU Facultad de Ciencia y Tecnolog?a Apdo. 644 48080 Bilbao (Spain) Fax: +34 94 601 3500 Tlf: +34 94 601 5437 ---------------------------------------- "UNIX was not designed to stop its users from doing stupid things, as that would also stop them from doing clever things." Doug Gwyn
