[Ocfs2-devel] flush

[Ocfs2-devel] flush

[Wim Coekaerts]

for those that are going to be awake now ;-)

sys_close() -> filp_close() -> ocfs_fops() -> ocfs_flush() -> fsync_buffers_list()

that happens on close. flush causes a sync to disk
so

untar of a full tree took 6:15 min
then took out the ocfs_sync_inode() from ocfs_file_release()
then lowered heartbeat to 1 per second nad commit interval now at 5
seconds

then went down to bout 4:30 minutes

then noticed the flush during untar, so I got rid of ocfs_flush() as a
test
went down to bout 2 minutes, however things are inconsistent
so it's not tha simple .but that's a huge part of the remaining
bottlenecks when comparing stuff to ext3

[Ocfs2-devel] flush

[Ling, Xiaofeng]

So there are two call to ocfs_sync_inode when closing file.
How about also removing the ocfs_sync_inode in ocfs_file_open ?

>-----Original Message-----
>From: ocfs2-devel-bounces@oss.oracle.com
>[mailto:ocfs2-devel-bounces@oss.oracle.com] On Behalf Of Wim Coekaerts
>Sent: 2004=C4=EA7=D4=C27=C8=D5 16:35
>To: ocfs2-devel@oss.oracle.com
>Subject: [Ocfs2-devel] flush
>
>for those that are going to be awake now ;-)
>
>sys_close() -> filp_close() -> ocfs_fops() -> ocfs_flush() ->
>fsync_buffers_list()
>
>that happens on close. flush causes a sync to disk
>so
>
>untar of a full tree took 6:15 min
>then took out the ocfs_sync_inode() from ocfs_file_release()
>then lowered heartbeat to 1 per second nad commit interval now at 5
>seconds
>
>then went down to bout 4:30 minutes
>
>then noticed the flush during untar, so I got rid of ocfs_flush() as a
>test
>went down to bout 2 minutes, however things are inconsistent
>so it's not tha simple .but that's a huge part of the remaining
>bottlenecks when comparing stuff to ext3
>
>
>
>_______________________________________________
>Ocfs2-devel mailing list
>Ocfs2-devel@oss.oracle.com
>http://oss.oracle.com/mailman/listinfo/ocfs2-devel
>
>=20

[Ocfs2-devel] flush

[Wim Coekaerts]

sure - but I don't think that matters any, at least not on this test.

On Wed, Jul 07, 2004 at 05:47:42PM +0800, Ling, Xiaofeng wrote:
> So there are two call to ocfs_sync_inode when closing file.
> How about also removing the ocfs_sync_inode in ocfs_file_open ?
> 
> >-----Original Message-----
> >From: ocfs2-devel-bounces@oss.oracle.com
> >[mailto:ocfs2-devel-bounces@oss.oracle.com] On Behalf Of Wim Coekaerts
> >Sent: 2004??7??7?? 16:35
> >To: ocfs2-devel@oss.oracle.com
> >Subject: [Ocfs2-devel] flush
> >
> >for those that are going to be awake now ;-)
> >
> >sys_close() -> filp_close() -> ocfs_fops() -> ocfs_flush() ->
> >fsync_buffers_list()
> >
> >that happens on close. flush causes a sync to disk
> >so
> >
> >untar of a full tree took 6:15 min
> >then took out the ocfs_sync_inode() from ocfs_file_release()
> >then lowered heartbeat to 1 per second nad commit interval now at 5
> >seconds
> >
> >then went down to bout 4:30 minutes
> >
> >then noticed the flush during untar, so I got rid of ocfs_flush() as a
> >test
> >went down to bout 2 minutes, however things are inconsistent
> >so it's not tha simple .but that's a huge part of the remaining
> >bottlenecks when comparing stuff to ext3
> >
> >
> >
> >_______________________________________________
> >Ocfs2-devel mailing list
> >Ocfs2-devel@oss.oracle.com
> >http://oss.oracle.com/mailman/listinfo/ocfs2-devel
> >
> > 
>