From: Joel Becker Date: Mon, 22 Mar 2010 13:41:59 -0700 Subject: [Ocfs2-devel] [PATCH 3/5] ocfs2: use allocation reservations for directory data In-Reply-To: <20100321232623.GN11402@wotan.suse.de> References: <1268809154-1020-1-git-send-email-mfasheh@suse.com> <1268809154-1020-4-git-send-email-mfasheh@suse.com> <20100319224310.GF15539@mail.oracle.com> <20100320001425.GL11402@wotan.suse.de> <20100320012554.GI15539@mail.oracle.com> <20100320034758.GM11402@wotan.suse.de> <20100320061847.GA14324@mail.oracle.com> <20100321232623.GN11402@wotan.suse.de> Message-ID: <20100322204158.GB14285@mail.oracle.com> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: ocfs2-devel@oss.oracle.com On Sun, Mar 21, 2010 at 04:26:23PM -0700, Mark Fasheh wrote: > On Fri, Mar 19, 2010 at 11:18:48PM -0700, Joel Becker wrote: > Aside from cleaning up the license info, I remove the dir_resv/no_dir_resv > options. I also added a flags field to struct ocfs2_alloc_reservation which > saves us the 'tmpwindow' argument to ocfs2_resmap_resv_bits(). I like the changes. > > I was thinking about a kernel tree untar where the directories > > are holding on to reservations as thousands of files are created under > > them. Wouldn't that lead to cannibalization and show us that pattern? > > This is just a lay guess - you have a lot more familiarity with the > > code. > > I don't know how much a kernel build is going to make a difference. I took a > couple images after kernel builds with various options but didn't see > anything obvious. To be fair though, I only checked one or two files. I'll > upload the images somewhere shortly. > > As an interim compromise, I changed the code to get minimum (8 bits) sized > windows on directories. That way, they'll get some amount of > continguousness, but not as much as file data. We can easily adjust in any > direction we want. I wonder how the varying reservation sizes will impact ulilization of the localalloc. May not matter, but there's a part of me that wonders. Does the new dir-resv code do as well on your tests as the no-dir-resv code? Maybe something to see any help from dir reservations would be a many-parallel-untar sort of thing? Unpack 20 kernel trees at the same time? Some other package with a lot of files. I have no idea if this will mean anything, and I actually rather trust your observations of previous untars. I'm just casting about for something to show us how dir reservations are behaving. Joel -- Life's Little Instruction Book #451 "Don't be afraid to say, 'I'm sorry.'" Joel Becker Principal Software Developer Oracle E-mail: joel.becker at oracle.com Phone: (650) 506-8127