On my linux servers I've been experiencing for a while with Aufs (a
"translucent" filesystem initially designed for applications like
Live-CDs) instead of using Raid or LVM to create multi-drive volumes.
Premises for this:
- raid is no backup, so you need a backup anyway,
- filesystem merging is more fault-tolerant than device merging: every
drive can fail independently, data on each drive can be
accessed/repaired separately,
- Buying drives when one needs replacement instead of all at a the
same time (for a Raid array) gives you more storage space,
- Drive performance is irrelevant: local access for the DB is always
fast enough, remote access speed is always limited by the network,
- I am lazy and don't like having to rush for repairs.
Aufs v1, in my experience, works great at balancing drive usage when
you add files to the drive pool, but doesn't work too well when you move
data around. If the pool is read-only, then it is quite reliable.
I've recently stumbled upon Mddhfs, a unioning filesystem for FUSE.
This one doesn't balance drive usage, but works well when you reorganize
files inside the pool. I currently use it for a read-write setup, and it
seems reliable.
This is an excerpt of my fstab file:
Code:
--------------------
LABEL=090824-DATA1 /mnt/data1 ext3
user,suid,nodev,exec,noatime,commit=600 0 2
LABEL=090824-DATA2 /mnt/data2 ext3
user,suid,nodev,exec,noatime,commit=600 0 2
LABEL=090824-DATA3 /mnt/data3 ext3
user,suid,nodev,exec,noatime,commit=600 0 2
LABEL=090824-DATA4 /mnt/data4 ext3
user,suid,nodev,exec,noatime,commit=600 0 2
mhddfs#/mnt/data1,/mnt/data2,/mnt/data3,/mnt/data4 /mnt/U-data fuse
nonempty,allow_other,mlimit=4G 0 0
/mnt/U-data/export /var/data/export none bind 0 0
--------------------
Aufs or Mhddfs group together files and subdirectories which, on the
different drives, live in a directory called the same. The root of each
drive looks like this:
Code:
--------------------
ls /mnt/data1/
anonftp export lost+found
--------------------
My SBS server points to /var/data/export/Audiofiles/. I have file
servers showing the /var/data/export hierarchy to client machines.
The total available space is close to the sum of available space on
each drive. Performance is not good, but in our case that's adequate.
Reliability. Let's say drive #2 and #4 fail or are gone for some
reason, you just need to change the mount to
mhddfs#/mnt/data1,/mnt/data3 /mnt/U-data ... and you're back in
business with the remainder of your files. SBS will need an update
rescan.
When you want to reorganize your file hierarchy, impacting the location
of, say, thousands of files, then it is probably better (speed,
reliability) to operate directly on each drive. That's straightforward
with mhddfs: "cd /mnt/data4" and do your stuff.
There are quirks to translucent filesystems, for example support for
hard-links, amount of free space reported, etc. so you need to try and
see for yourself. I think Mhddfs can of use to some Squeezebox users.
http://debaday.debian.net/index.php?s=mhddfs
http://mhddfs.uvw.ru/
--
epoch1970
------------------------------------------------------------------------
epoch1970's Profile: http://forums.slimdevices.com/member.php?userid=16711
View this thread: http://forums.slimdevices.com/showthread.php?t=70287
_______________________________________________
unix mailing list
[email protected]
http://lists.slimdevices.com/mailman/listinfo/unix
