We must launch several hungry processes on a Linux system. These processes often need several Go (~5Go) of memory to run (total memory: 16Go RAM + 2Go swap).
At first, when the system was running out of memory, the OOM-killer killed processes, we had to reboot the system each time it happened.
Then, we have tried to use the overcommit_memory (= 2) + overcommit_ratio (= 75) parameters, so processes are not launched when the situation is getting critical. So, no need to reboot the server anymore. But our launching script now report dozens of errors when the limit is reached : the new processes are immediately in error, the processes are never launched.
So now we are looking for a solution to launch "as many" processes as we want, and then they would be delayed/paused or anything, waiting for their brothers to stop... Does it exist ?