Re: [PATCH 0/3 v3] dcache: make it more scalable on large system

["Jörn Engel" <joern@logfs.org>]

On Thu, 30 May 2013 11:48:50 -0400, Waiman Long wrote:
> On 05/29/2013 05:19 PM, Jörn Engel wrote:
> >On Wed, 29 May 2013 22:37:00 +0200, Andi Kleen wrote:
> >>>As Dave said before, is the last path component sufficient?  Or how
> >>>about an inode number?
> >>Neither works, the profiler needs to find the file and read it.
> >Ignoring all the complexity this would cause downstream, you could do
> >the path lookup just once, attach some cookie to it and return the
> >cookie ever-after.  Maybe some combination of i_sb and i_ino would be
> >good enough as a cookie.
> 
> Still, it is just shifting the complexity from the d_path code to
> the perf kernel subsystem as it needs to keep track of what paths
> have been sent up before.

That sounds like a good thing to have.  Every single linux user
depends on the dcache.  Only a relatively small subset cares about
perf.  Having dcache pay the cost for perf's special needs is a
classical externality.

> It also have complications in case the
> tracked files are being deleted or moved around in the filesystem.
> Some kind of notification mechanism has to be implemented in the
> dentry layer to notify the perf subsystem.

Agreed.  The whole approach is based on getting the 99% case right and
occasionally being wrong.  For perf this may be acceptable, not sure.

Jörn

--
Without a major sea change, nothing that is under copyright today will
ever come out from under it and fall into the public domain.
-- Jake Edge
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html