Re: [PATCH] writeback: Fix broken sync writeback

[tytso@mit.edu]

On Tue, Feb 23, 2010 at 01:53:50PM +1100, Dave Chinner wrote:
> 
> Ignoring nr_to_write completely can lead to issues like we used to
> have with XFS - it would write an entire extent (8GB) at a time and
> starve all other writeback. Those starvation problems - which were
> very obvious on NFS servers - went away when we trimmed back the
> amount to write in a single pass to saner amounts...

How do you determine what a "sane amount" is?  Is it something that is
determined dynamically, or is it a hard-coded or manually tuned value?

> As to a generic solution, why do you think I've been advocating
> separate per-sb data sync and inode writeback methods that separate
> data writeback from inode writeback for so long? ;)

Heh.

> > This is done to avoid a lock inversion, and so this is an
> > ext4-specific thing (at least I don't think XFS's delayed allocation
> > has this misfeature).
> 
> Not that I know of, but then again I don't know what inversion ext4
> is trying to avoid. Can you describe the inversion, Ted?

The locking order is journal_start_handle (starting a micro
transaction in jbd) -> lock_page.  A more detailed description of why
this locking order is non-trivial for us to fix in ext4 can be found
in the description of commit f0e6c985.

Regards,

							- Ted