On Wed 11-07-12 12:07:26, Jan Kara wrote: > On Wed 11-07-12 12:58:16, Artem Bityutskiy wrote: ... > > And this is the behavior this patch modifies: we stop using 's_dirt' and just > > mark the superblock buffer as dirty right away. Indeed: > > > > 1. It does not add any value to delay the I/O submission for cases 1-3 above. > > They are rare. > > 2. Case number 4 above depends on whether we have file-system checksumming > > enabled or disables. > > a) If it is disabled (most common scenario), then it is all-right to just > > mark the superblock buffer as dirty right away and it should affect > > performance. > > b) If it is enabled, then we'll end up doing a bit more work on deletion > > because we'll re-calculate superblock checksum every time. > > > > So case 2.b is a bit controversial, but I think it is acceptable. After all, by > > enabling checksumming we already sign up for paying the price of calculating > > it. The way to improve checksumming performance globally would be to calculate > > it just before sending buffers to the I/O queue. We'd need some kind of > > call-back which could be registered by file-systems. Actually, the most common case of adding orphan inode used ext4_handle_dirty_super_now() so for that case there is no difference. And other cases are so rare it really does not matter... So there shouldn't be any measurable difference. Honza -- Jan Kara <jack@xxxxxxx> SUSE Labs, CR -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
