[Wylug-discuss] Help needed with 'failed' Linux software RAID 10

[John Hodrien]

On Sat, 26 Jun 2010, Dave Fisher wrote:

> Since I have 4x2TB and 8x1TB discs, I was thinking of putting 2x2TB discs and
> 3x1TB drives in each barebones server ... and dividing every disc into 100GB
> partitions.
>
> If I'm not mistaken, that would allow me to construct 10x100GB RAID-5 arrays
> and 10x100GB RAID-1 arrays in the 5-bay Tranquil PC BBS2 enclosure.

> Which would yeild 4TB of RAID-5 storage and 1TB of RAID-1 storage, with all
> 20 arrays able to survive (at least) one disc failure.

Yes, I guess you could.

> Although 100GB partitions 'feel' a little tight for video work (especially
> 'raw' DV/HD), I picked this number with the following in mind:

Won't your RAID5 partitions be 400Gb each (assuming that you're not then
subdividing these again)?

> Given the relative performance, I'm guessing that active data should go on
> the RAID-1 arrays and archive data on the RAID-5 arrays.

Don't be so certain about these figures.  You'd be comparing a 5 disk RAID5
with a 2 disk RAID 1.  Read performance should be a shoe in for the RAID5.
Write performance could still be faster on the RAID5.  In either case, across
a gigabit network with few concurrent read/write threads, you could easily be
gigabit limited with a single disk.

> Having reflected on John H's defence of LVM, I think I might use it for
> snapshotting active data and systems files. However, I am not sure what
> benefits it might offer for managing archive data.

My view of LVM is that its benefits often come into play when you least expect
it, later down the line.  Especially related to disk failures and upgrades,
there have been times where I've mourned not installing with LVM in the first
place, yet there haven't been times where I've wished the opposite.

>  e. Although the RAID-5 recovery time would be no better than could be
>  achieved with 200GB partitions, I'm guessing that it won't be much worse
>  (assuming parity is distributed 'round-robin').

It's RAID5 not RAID4.

>  f. Although recovery of a RAID-1 mirror from complete disc failure will
>  still be time-consuming, I'm guessing that recovery from badblock/metadata
>  errors within just one 100GB partition should be a lot quicker (assuming
>  that you can fail a partition-based array member, without failing the
>  entire physical disc).

Yes, although once you've got RAID issues caused by bad blocks, you're minutes
away from ordering a replacement disk anyway.

jh