<div class="gmail_quote">On Thu, Nov 3, 2011 at 5:23 AM, Lawrence Rust <span dir="ltr"><<a href="mailto:lvr@softsystem.co.uk">lvr@softsystem.co.uk</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
I don't know if anyone else has tried the Linux 3.1 kernel but I have a<br>
cautionary note if you use an ext3 partition. Linux 3.1 will ship with<br>
Ubuntu 12.04 and Fedora 16. It comes with ext3 barriers enabled by<br>
default, which has a dramatic negative effect on file system write<br>
performance. For those unfamiliar with ext3 barriers see this link:<br>
<a href="http://lwn.net/Articles/283161/" target="_blank">http://lwn.net/Articles/283161/</a><br>
<br>
I still use an ext3 partition because of the poor performance using<br>
mysql with ext4:<br>
<a href="http://www.tedman.com/techie-notes/fedora12onthinkpadx61s" target="_blank">http://www.tedman.com/techie-notes/fedora12onthinkpadx61s</a><br>
<br>
Ext3 barriers have been around since Linux 2.6.9. but have been disabled<br>
by default. Some distros, like Red Hat, have enabled them but most<br>
don't. With Linux 3.1 barriers are enabled by default which can improve<br>
file system journal robustness in the event of power failures but at the<br>
expense of significant disk cache performance loss. I noticed at least<br>
twice the disk activity and a 10% increase in cpu usage while just<br>
recording SD.<br></blockquote></div><br>My first run-in with barriers was with ext4 vs. ext3 while running the Folding@Home client on various machines. The F@H client uses some sort of synchronous writes when it's done with the work units. Between each work unit on ext4 it would take 40-60 minutes before moving onto the next unit. On ext3 is was less than 15 minutes. When I enabled barrier=0 on ext4, the performance improved tremendously and I understand the consequences should I lose power. :-)<br>
<br>/Brian/<br>