shmmax shmall still unclear
"still unclear" means to me. For a computational code, i need
allocating "replicated memory" (i.e., the amount of memory for each
CPU) and "distributed memory" (i.e., the memory allocated for the
node) in a UMA-type computer of four dual-opterons with 24GB physical
ram (amd6s lenny). A correct allocation is very stringent, as the
procedure checks for needed memory for a given job and does not start
if memory is not enough (big matrices to solve).
I know how to set shmmax and shmall through /etc/sysctl.conf and i
found so much on internet about setting these parameters for
databases, though it is unclear to me how to do at the best in the
general context of number crunching in solving matrices, as it occurs
with computational codes like, for example, gamess-us. Currently the
setting is shmmax=shmall=90% physical memory installed but surely the
two parameters should have their proper - and different - setting.