Re: trying to understand READ_META, READ_SYNC, WRITE_SYNC & co

[Christoph Hellwig <hch@lst.de>]

On Mon, Jun 21, 2010 at 08:58:51PM +0200, Jens Axboe wrote:
> It's definitely a win in some cases, as you showed there as well.
> My initial testing a long time ago had some nice benefits too. So
> perhaps the above wasn't worded very well, I always worry that we
> have regressions doing boosts for things like that. But given that
> meta data is something that needs to be done before we get to the
> real data, bumping priority generally seems like a good thing to do.

Even if the REQ_META special casing helps with performance it creates
a big issue if we want to follow your other guide line, that is marking
all actual metadata requests REQ_META for blocktrace.  What about
only applying the metadata preference only to _synchronous_ (read or
REQ_SYNC) I/Os that also have REQ_META set?

Right now we never use REQ_META on a non-synchronous request (XFS appears
to, but the code is not actually reachable anymore), so it's not
actually a change in behaviour.  After that we could do an easy sweep
through the tree and mark all metadata requests as REQ_META.  Btw, what
do we consider metadata for this purpose?  The interesting question
here is about indirect blocks / bmap btree blocks.  In the traditional
sense they are metadata, but for I/O purposes they are mostly part of
the I/O stream.