Ext3 users - Apr 2002 - writing processes are blocking in log_wait_common with data=ordered

I have a system with many processes writing to a common data file using 
pwrite.  These processes are each writing to existing blocks in the file, 
not changing the file size, and the file has no holes.

When the processes get going, they seem to bottleneck at log_wait_common 
(according to ps alnx).  That is, one process is uninterruptible in
log_wait_common, the rest are uninterruptible in down.

kjournald itself seems to be oftern blocked at 'wait_o', which I
unfortunately didn't get as numeric, but I assume is wait_on_buffer or
something like that.

Is there a reason the data writes should be waiting on the log in ordered 
mode?  I thought that log commits were postponed until the completion of 
data, but not vice-versa.

I assume that data=writeback would alleviate this bottleneck, right?

David

-- 
/==============================\
| David Mansfield              |
| david@cobite.com             |
\==============================/

I hate when this happens, but I have to reply to myself.  Of course, the 
subject line should read: log_wait_commit.  And I should mention that the 
file is opened with O_SYNC.

This info should help.

David

-- 
/==============================\
| David Mansfield              |
| david@cobite.com             |
\==============================/

On Apr 30, 2002  14:58 -0400, David Mansfield wrote:
> I hate when this happens, but I have to reply to myself.  Of course, the 
> subject line should read: log_wait_commit.  And I should mention that the 
> file is opened with O_SYNC.
Well, if you are using O_SYNC, you are probably better off having a
large journal and using data=journal mode.  Otherwise I would have
suggested data=writeback because this is best for databases.

Cheers, Andreas
--
Andreas Dilger
http://www-mddsp.enel.ucalgary.ca/People/adilger/
http://sourceforge.net/projects/ext2resize/