Friday, January 29, 2010

My beef with Linux RAID

I have used software RAID in Linux in the past, via mdadm. I believe it has resulted in lost data more times than it has saved me from losing data. I have since switched to ZFS (on FreeBSD) and have survived many situations that could have resulted in lost data.

I'd like to be able to switch back to using Linux, so I have been doing some research on Linux RAID. Today I came across this article, which contains the following text:

For a RAID-1 config with two disks...If both blocks were readable (i.e. were read from the disk, without the disk indicating any sort of error condition), but there was a mis-compare, then the data from the highest-numbered disk is copied to the other disk. This results in a 50-50 chance that good data was over-written by bad. Furthermore, this is done silently: no syslog messages indicate either a mis-compare, or that a repair action was taken.
That's right! mdadm doesn't checksum your data to prevent this from happening! ZFS does. So, for example, you could write random data to one of your drives:

cat /dev/random > /dev/hda1

In ZFS, no problem! In RAID-1.... fail?

Reblog this post [with Zemanta]

1 comment:

  1. This comment has been removed by a blog administrator.

    ReplyDelete