Monique Y. Herman wrote:
On Fri, 10 Oct 2003 at 11:09 GMT, Tim Connors penned:Not a case of ext3 being crap, a case of ext3 with journalled *data* being crap. Quite a nice allrounder with the other two ext3 options set. And you get the same problems with all other fses when their equivalent of journalled *data* was turned on (if they had such a feature).I read that but didn't understand it. Is it that you can use ext3 without journalling? Or is journalling data different from normal journalling somehow? I'm confused.
I believe they are referring to the type of journaling being done. The default on Woody 2.4.18-bf2.4 (and RH7.3) is data=ordered. With the first version of ext3 or if you are using data=journal, the data is written to the journal and then to the normal location on the file system. With data=ordered only metadata is written to the journal but it guarentees that it won't commit transactions until the real data has hit the disk.
This is a pretty good EXT3 faq http://batleth.sapienti-sat.org/projects/FAQs/ext3-faq.htmlIf I am rememering correctly other journaling file systems journal metadata as well. It is obviously a larger performance hit to write the same data twice; data=ordered avoids that and still gives good journaling protection. As fhe FAQ points out, you could use data=writeback for even less of a runtime performance hit with a faster fsck recovery than ext2.
It was my understanding that ext3 is ext2 with some additional structure information (within the space allocated for such things by ext2, so 100% reverse compatable) and a journal file. I hadn't heard anything about needing to be careful about mounting rw with a "dirty journal" before this thread. That is something that I'll read into.
I've been using ext3 since it shipped standard on RH with version2 data=ordered. That's been at least a year and a half (probably longer) and I've never had a problem on the dozen machines I've run it on. Ext2, while not extreme in any performance spec, has been a reliable Linux FS, and ext3 just builds onto that. I don't loose any sleep over any data I put on it, and I have yet to fret over how it affects dsk I/O on any of the servers.
-- Jacob