XFS, ext and per-inode mutexes [by Mark Callaghan]

I still don't think this is widely known. It should be. ext-2 and ext-3 lock a per-inode mutex for the duration of a write. This means that ext-2 and ext-3 do not allow concurrent writes to a file and that can prevent you from getting the write throughput you expect when you stripe a file over several disks with RAID. XFS does not do this which is one of the reasons many people prefer XFS for InnoDB. I don't know whether ext-4 does this.
Depending on what caches are enabled in your storage, this can have reduce disk write throughput. Even when you have a battery backed write cache, unless the batteries can be hot-swapped, it is likely that at some point in time your server will run in production with the RAID write cache disabled.
I have a test server that uses HW RAID 10 with disk caches disabled. I disabled the battery backed write cache on HW RAID and ran sysbench fileio tests using ext-3 and XFS. The server uses CentOS 5.2. The numbers below repeat results I gathered a few years ago after browsing ext-2 source code to confirm the behavior.
Writes-per-second
threads 1 2 4 8
xfs 162 294 501 800
ext-3 161 162 161 165

Notes

Script to run sysbench. It was run for 1, 2, 4 and 8 threads for 30 seconds.

nt=$1
secs=$2
args=$3
./sysbench --test=fileio --file-num=1 --file-total-size=4GB \
--file-test-mode=rndwr --max-requests=0  \
--file-fsync-freq=0 --file-extra-flags=direct \
--num-threads=${nt} --max-time=${secs} ${args}

http://www.facebook.com/notes/mark-callaghan/xfs-ext-and-per-inode-mutexes/10150210901610933