[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20121113152846.GJ7808@moon>
Date: Tue, 13 Nov 2012 19:28:46 +0400
From: Cyrill Gorcunov <gorcunov@...nvz.org>
To: Tvrtko Ursulin <tvrtko.ursulin@...lan.co.uk>
Cc: David Rientjes <rientjes@...gle.com>,
Andrew Morton <akpm@...ux-foundation.org>,
linux-kernel@...r.kernel.org, linux-fsdevel@...r.kernel.org,
Al Viro <viro@...iv.linux.org.uk>,
Alexey Dobriyan <adobriyan@...il.com>,
Pavel Emelyanov <xemul@...allels.com>,
James Bottomley <jbottomley@...allels.com>,
Matthew Helsley <matt.helsley@...il.com>,
aneesh.kumar@...ux.vnet.ibm.com, bfields@...ldses.org
Subject: Re: [patch 3/7] fs, notify: Add file handle entry into
inotify_inode_mark
On Tue, Nov 13, 2012 at 03:02:22PM +0000, Tvrtko Ursulin wrote:
> On Tuesday 13 November 2012 18:40:36 Cyrill Gorcunov wrote:
> > On Tue, Nov 13, 2012 at 12:37:23PM +0000, Tvrtko Ursulin wrote:
> > >> Which would give about 26K of additional memory if c/r get used here.
> > >>
> > >> Not a big number i guess?
> > >
> > > I am pretty sure there are desktop file indexing packages which use
> > > inotify or fanotify which will put a mark on every single directory within
> > > users home.
> > >
> > > You probably need to test this with default installs of popular desktop
> > > environments and realistic home directories.
> >
> > I'm about to shrink the handle down to 40/64 bytes as being proposed in one
> > of early review cycles (i'll do that with patch on top), which should
> > minimize the amount of memory needed (look, it's pretty clear that if the
> > system uses millions of inotify watchers each inotify mark will need the
> > fhandle here in c/r sake, i simply see no way at moment how to escape this
>
> Well I spotted uncertainty in this thread about how many of these structures
> will typically be instantiated at runtime which is what I tried to add to this
> discussion. I have 60k directories in my home for example...
And I appreciate it, really, thanks for info!
> If you don't want to use information I provided that is your choice. Because
> you are still probably doubling this structure. Give or take - I haven't
> actually bothered counting.
>
> Perhaps there could be a different way, where you could use additional space
> only when it is actually used at runtime. But as I said, I am not following
> closely.
Unfortunatelly at moment I see no way how to make it in on-demand fashion.
>
> > completely, but if the c/r is turned off, which is by default, no
> > additional memory needed).
>
> By turned on and off you are not talking about runtime but about kernel
> compile time, right? Do you envisage distributions turning this on, like they
> do for most things?
yeah, compile time only. i don't expect it to be turned on by default
sometime soon but sure distro people have own opinions.
Cyrill
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists