On 08/02/18 12:56, Phil Turmel wrote:
On 02/07/2018 10:14 PM, NeilBrown wrote:
On Thu, Feb 08 2018, Wol's lists wrote:
I've been playing with a mirror setup, and if we have two mirrors, we
can rebuild any failed disk by coping from two other drives. I think
also (I haven't looked at it) that you could do a fast rebuild without
impacting other users of the system too much provided you don't swamp
i/o bandwidth, as half of the requests for data on the three drives
being used for rebuilding could actually be satisfied from other drives.
I think that ends up being much the same result as a current raid10
where the number of copies doesn't divide the number of devices.
Reconstruction reads come from 2 different devices, and half the reads
that would go to them now go elsewhere.
This begs the question:
Why not just use the raid10,near striping algorithm? Say one wants
raid6 n=6 inside raid60 n=25. Use the raid10,near6 n=25 striping
algorithm, but within each near6 inner stripe place data and P and Q
using the existing raid6 rotation.
What is the more complex placement algorithm providing?
It came from the declustered thread.
Especially with raid-60, a rebuild will hammer a small subset of the
drives in the array. The idea is that a more complex algorithm will
spread the load across more drives. If your raid-6 in a raid-60 has say
8 drives, a rebuild will stress 16 drives. If you've got 100 drives
total, that's a lot of stress that could be avoided if the data could be
more widely spread.
Thing is, you CAN gain a lot from a complex raid like raid-60 which you
lose with a raid-6+0 - again something that came up was you have to
scrub a raid-6+0 as a whole bunch of separate arrays.
Really, it's a case of the more we can spread the data, it (1) reduces
the stress during a rebuild, thus reducing the risk of a second related
failure, and (2) it increases the chances of surviving a multiple drive
failure because if three logically related drives fail you've lost your
raid-6-based array. Spreading the data reduces the logical linking
between drives.
Cheers,
Wol
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
