In article < XXXX@XXXXX.COM >,
The difference is (almost exactly) 2GB extra "reserved" swap space.
This is indicative of something allocating a lot of virtual memory,
but not actually using it, e.g. a mammoth malloc() or an anonymous
mmap (without MAP_NORESERVE).
Try checking for processes with huge VSZ values, e.g. by
ps -eo pid,vsz,comm | sort +1n | tail
If you identify a likely culprit, apply pmap(1) to it.
Email: cet1 [at] cam.ac.uk