On 21.02.2012 09:46, babajaga wrote:
I switched from ufs to aufs, and recognized, that using cachemgr,
"ASYNC IO
Counters" always show
"write:0", athough the other counters for read, unlink etc. have high
values.
Digging into the source code, in /src/fs/aufs/store_asyncufs.h, I
found
#define ASYNC_WRITE 0
obviously disabling async writes.
Is it disabled because being broken/unstable, or any other reason ?
This is the reason we still say diskd is fastest on BSD but AUFS is
faster on Linux.
As I heard the story the write handlers are not designed to work
properly as async handlers yet. That could be wrong so you are welcome
to experiment, but I would advise against enabling it for production
before some good testing.
Patches on 3.2 or 3.HEAD/(3.3) are welcome. 2.7 is already scheduled
for deprecation with 3.2 release so patches there will not be of much
use.
HTH
Amos