[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: EXT3 with 2.2.20 - Is it stable enough for a production server th at is used 24X7?



Hi,

On Fri, Nov 16, 2001 at 01:00:37AM -0800, Mike Fedyk wrote:

> > Probably in 2.5, there will be changes to allow the VM to tell the
> > journal to free up some memory when it is running low.
 
> But, this is ext3 on a 2.2 kernel.  Doesn't 2.4 work better on this?

Yes.

The fundamental problem is that the 2.2 VM really has no way of
telling the filesystem when we are under VM pressure --- so ext3 can't
respond.  Under 2.4, the new "address_space" abstraction makes it all
much easier, and the ext3 patch adds a very small extension to the
address_space operations to let ext3 free the metadata it attaches to
existing fs structures on demand.

Fixing this on 2.2 requires substantial VM surgery and is not a high
priority.  One thing that could be done relatively simple (and Ted
Ts'o sent me patches to do this partially) would be to account the
ext3 VM usage and to throttle ext3 globally if that exceeds a certain
value.  That is _not_ a real solution: if you do things like
mlock()ing down some memory, ext3 won't be able to adapt to the new
free-space availability.  But it would work for most situations.

The real answer, though, is 2.4.  The ext3 performance on 2.4 has been
studied a lot and is _far_ better than on 2.2, and there's little
chance that all of that work will be backported since the 2.2 VM makes
ext3 look rather different on that kernel anyway.

Cheers,
 Stephen





[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]