On 2012-10-07 15:19, Swâmi Petaramesh wrote: > Hi again ;-) > Le 07/10/2012 14:33, Alex a écrit : >> 1. Convert to a 16k or 32k leafsize. > How should I do this ? Can I do this on a live FS, and isn't this going > to double my on-disk used space (I have active snapshots...) >> 2. defragment (each non-trivial file) every now and again > I believed that the "autodefrag" mount option was supposed to avoid the > need for manual defrags...? > > Plus, with respect to snapshots, isn't this going to increase a lot my > used disk space ? > >> 3. Note that btrfs in kernel 3.6 (vs 3.5) does appear to be quite a bit quicker. > I'm not sure that "a bit" quicker will be enough, when I see my BTRFS is > currently about 8x slower than what ext4 user to... > Here is a graph to reinforce what Alex said about performance after 3.5.x. This graph was posted in #btrfs on freenode. At the end of the graph, the poster said that he upgraded from a 3.5.x to a 3.6.x https://pipebreaker.pl/dump/loadl.png I noticed the same thing going from 3.5.0 to 3.7.0rc2; system usage is much lower. Before this I would often find huge lags and then check and see btrfs-cleaner, btrfs-transactionhelpersomething, etc. taking up 100% io. So if your system usage is high, this will help. Also possibly not relevant for you: on my desktop, (btrfs or ext4, with consumer disks) I find that deadline scheduler and nr_requests of 8 is vastly superior for reducing GUI lags. (default is scheduler cfq and 128 nr_requests). But for servers, I don't know what to recommend. -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
