I have noticed a big swapout performance regression on recent kernels using
the following testcase:
swapon /dev/md0 (raid0)
mount none /dev/shm -t tmpfs -o size=10G
cat /dev/full > /dev/shm/tst.dmp
Kernel starts to swapout at full speed after filling physical ram.
Then, kernel progressively slows down swapout after crossing the ~3G VM
threshold, compounding at over 50% slowdown when reaching the 10G tmpfs
limit.
Has this been fixed?
Thanks!
--
Al