Page 3 of 3

Re: WU's fail after 100%, left over work files, ext4 cause?

PostPosted: Sat Feb 13, 2010 4:11 am
by chrisretusn
Out of curiosity, why are your folding off a live CD via ram disk? At least that is what it looks like to me.

Re: WU's fail after 100%, left over work files, ext4 cause?

PostPosted: Tue Mar 16, 2010 12:12 am
by bruce
Ext4 is not the only issue here. Other filesystems and other OSs have similar features.

We're seeing an number of reports where checkpoints cannot be recovered after a sysem crash. It's very difficult to tell whether this is related to delayed allocation or delayed writes or neither. Certainly there is a tradeoff between file system speed and reliability. Consider the following
http://es.linuxfoundation.org/news-medi ... le-problem

If you or your system designer want to speed up your filesystem, there's no doubt that these techniques work, provided you don't really care what happens if the system crashes or you have to do a forced reboot when the file system is still dirty. If you think that FAH spends too much time waiting for the fsync() processes to finish, consider the alternatives.

Re: WU's fail after 100%, left over work files, ext4 cause?

PostPosted: Tue Mar 16, 2010 12:30 am
by chrisretusn
Agree. I spent considerable time testing this and results were mixed with ext4. While I could reproduce failures most of the time it was not 100% of the time. One thing for sure, since switching back to the jfs file system I have not had any more problems.

Re: WU's fail after 100%, left over work files, ext4 cause?

PostPosted: Tue Mar 16, 2010 1:40 am
by Wrish
Wow, I didn't realize the HFS+ system in OS X was merely exhibiting the delayed write behavior found in -ext4. That would explain why most people never see this issue - the ones running Windows with NTFS or Linux with -ext3, both with default settings.