[mythtv-users] Linux software raid question

[Yan-Fa Li]

On Tue, Jun 3, 2008 at 7:42 PM, Carl L. Gilbert
<clg-social at rigidsoftware.com> wrote:
> On Tue, 2008-06-03 at 17:41 +0200, Boleslaw Ciesielski wrote:
>> Matt Nelson wrote:
>> > On one of my mythtv backend servers I host my fileserver that stores all
>> > of my media on two raid5 sets totaling 12 drives that are all attached
>> > via sata.  These are all attached externally, and that is where my
>> > question lies, if I connect these drives incorrectly will my raid sets
>> > die, or does it not matter where they are connected since there is some
>> > metadata that tells the linux software raid that it belongs to a certain
>> > raid set?
>>
>> Yes, Linux software RAID will work regardless of the drive order. Your
>> only order concern is the boot drive.
>>
>
> No, this is not guaranteed.  It depends on the distro and how the drives
> are referenced.  I think more recent distros should work.  But I know
> for a fact that older ones would not.  Like old Redhat versions where I
> first cut my teeth on RAID.
>
> Anyway, just step up to hardware RAID and don't even think about it
> anymore.
>

mdadm based raid is extremely reliable if you're using UUIDs.
Hardware RAID is not only expensive but it doesn't scale as well as
software RAID.  Here's a great blog entry by Jeff Bonwick one of the
architects of Sun's ZFS.  He has a lot of great points and confirms a
lot of my own thoughts about what to use all those cores on new CPUs
for.  http://blogs.sun.com/bonwick/entry/the_general_purpose_storage_revolution.

There are two large drawbacks of hardware RAID.  One is that you can't
transport RAID disks between different manufacturers RAID cards.
You're locked into to a specific vendor.  With Linux soft RAID, I just
move the disks over to a new machine and off I go.  The other is the
inability to run smartmontools and get at least a chance 60% chance of
spotting a bad disk.  Most hardware RAID implemtations hide the low
level block devices away from the operating system and do not give you
access to the individual smart information per drive unless you run a
vendor specific tool.

When one adds in the cost of hardware RAID, and any potential upgrades
like battery backed RAM caches, I would much rather put that money
toward a server quality motherboard with ECC RAM, a largish UPS unit
and set up RAID drives with write-intent bitmaps[3] and do weekly data
check sweeps[1] via cron combined with smartmon short and long disk
checks[2] on a daily and weekly basis.  Most of the features of the
high end harware RAID cards are now part of the feature set of Linux
soft RAID, it really is that good.

Yan

[1] e.g. echo "check" >> /sys/block/md0/md/sync_action
[2] e.g. smartctl -t short|long /dev/sda
[3] e.g. mdadm /dev/md0 -Gb internal