>> I'd like to know if anyone here has any thoughts or opinions about the >> best linux filesystem to use for an email system. There will be some >> small amount of website data on the system (including webmail to read >> the emails), although I could move that to another partition if need >> be. >> >> Anyone use ext4? Btrfs? Something else? Is ext3 fine even in high >> volume email servers? Anyone use any non-default values for block size >> or journaling type? > > We are currently testing XFS 3.1.1 on Centos 6.2. We got the knack for > XFS after watching this video where an XFS dev speaks to the new features > and other interesting aspects of filesystem development: > > http://www.youtube.com/watch?feature=player_embedded&v=FegjLbCnoBw
Thank you for the link. > Namely the way they reduce metadata bandwidth was a huge improvement for > me. Per Stan's response on this thread, what kernel does CentOS 6 use nowadays? (I know RedHat tends to maintain slightly older versions of things, although I think they backport changes from newer versions (making it even harder to know what their version numbers really mean)) > In our environment we have 2.5 TB of maildir structured files, and moving > it to XFS (from ext3) has been like night and day. We had a script that > created all the user's home directories on the new filesystem in this > format '/mail/a/n/andy/Maildir/blah', for 65000 user accounts. It > completed in about five minutes, which ext3 had taken upwards of two hours > to do in the past. Big difference. We then started moving the mail over > to the new file system using 4 parallel rsyncs it took about 2 hours to > move the mail to the new file system, which was fairly quick. So far > everything seems a bit snappier when loading mail into a client, or into > webmail. The mail servers are running low load average, and the IOPS have > decreased by about a fifth from the ext3. Very promising. Is it only your maildirs on XFS? Are the mounts to local drives? > Now if you're looking at doing something a bit more extravagent you should > look at GlusterFS, a distributed file system that can write synchronously > mirrored copies of files to multiple storage nodes, and exports it's file > system using NFS as well as an improved client provided by Gluster that has > better file locking. You can use XFS/EXT4/BTRFS/etc. underneath GlusterFS > as your core storage, and the GlusterFS runs on top of the multiple nodes, > distributing files over the cluster making a highly performant and highly > avialiable storage backend. We have done quite a bit of testing with this, > and there are certainly some gotchas to be aware of when storage nodes > fail, but it's well documented. Overall it's a great tool if you are > looking at doing high demand file serving and could benefit from a > clustered file system. I don't think that's where we're headed, but I do appreciate the tips.
