Michael Stahl <mstahl(a)redhat.com> writes:
what's the point of putting swap onto a hard disk in the first
place,
with today's RAM sizes (or even 10 years ago) of GBs, when swap is read
in unpredictable random order that incurs 10ms of latency each time?
It allows you to run in strict overcommit mode (with
vm.overcommit_memory=20). Swap is only there as a fallback, to allow the
kernel to increase its commit limit. It is basically never used.
Strict overcommit helps to salvage the system in case there are multiple
run-away processes allocating lots of memory.
Florian