Hi,
I tested setting up a swapfile on an md device: ( 2.0.36+0145raid)
I created an 8MB swapfile on the md device,
dd if=/dev/zero of=/swapfile bs=1024 count=8192
mkswap /swapfile 8192
swapon /swapfile
at this point the swapon process hangs and says:
"Got md request, not good ..."
even trying to kill -9 the swapon process does not works,
the swapon program hangs ...
I think the problem is the thing allan pointed out:
when using a swapfile the kernel bypasses the filesystem and uses the
blocks on the disk directly,
which fails in the case of soft-RAID disks, because the blocks are
spread across disks ....
Some time ago I patched an old 2.0.12 kernel with the swap-over-NFS
patch, for a diskless
machine and it worked very well.
I think swap over NFS must go through the filesystem of the NFS server,
because the NFS server
does not permit to use the disk blocks directly.
Does anyone know if there is a similar pacht that allows to swap over a
soft RAID5 disk ?
Or is there a way to declare part of the RAM as swaparea ,
maybe using /dev/loop ?
this because linux without a swaparea , has big performance penalties
when running out of RAM
(due to executable pages trashing).
thans for infos,
regrads,
Benno.
