This is an old benchmark, but <a href="http://www.phoronix.com/scan.php?page=article&item=linux2635_btrfs&num=1" rel="nofollow">http://www.phoronix.com/scan.php?page=article&item=linux...</a> (June 2010) shows BTRFS still has some weird performance problems - eg with Apache, 400 req/second vs 12000+ req/sec for EXT4.<p>The big problem with Linux filesystem benchmarks is that the filesystems are so tuneable the conclusions can often be better stated as "it is possible to tune file system X to perform badly" rather than "X is faster than Y".
It's hard to get a handle on the API usage mix of the various synthetic and app benchmarks described here. Some analysis breaking things down and revealing the performance model behind the FSes would be good, so that particular benchmark performances could be predicted from the model.<p>For example, SQLite is known to use fsync, and btrfs fared poorly there, so can we infer that btrfs doesn't handle fsync well with a particular pattern of writes / etc. beforehand?
I actually spent some time using NILFS2 as a root filesystem on an SSD, but I could never get it to work properly, mostly due to lack of documentation and the added complexity of using a daemon to manage your filesystem.<p>It'd be nice to see a write up on the current state of BtrFS from someone involved in the project.