(First time poster, so please be gentle.
)
One of the things that I've seen in various MythTV docs, including those for KnoppMyth, is "Don't use ext3 for the cache file system". The implication given in most of these documents is that ext2 or xfs performs acceptibly for the LiveTV cache, but ext3 does not. Has anybody tried to quantify this?
Here's why I ask: As part of my day job, I do a lot of work with parallel file systems, particularly PVFS and more recently PVFS2. Both of these put their "back end" on local file systems, and as with MythTV's LiveTV cache, we saw worse performance with ext3 than we did with ext2 or xfs.
Then we monkeyed with the ext3 mount options.
It turns out that the default mount options for ext3 include "data=ordered", which requires data to be flushed disk before updating the journal for the metadata. OTOH, xfs journals metadata without requiring a data flush, and of course ext2 doesn't journal anything at all. It is possible to change ext3's journaling behavior (and performance) to be similar to that of xfs by using the mount option "data=writeback".
Has anyone experimented with using ext3 as the cache file system with this more permissive journalling option? (I've converted the cache FS on my KnoppMyth box to xfs, so I haven't had a chance to try this myself yet.)