Re: [LSF/MM TOPIC] De-clustered RAID with MD

[NeilBrown <neilb@suse.com>]

[-- Attachment #1: Type: text/plain, Size: 1848 bytes --]

On Tue, Jan 30 2018, Wols Lists wrote:

> On 29/01/18 21:50, NeilBrown wrote:
>> By doing declustered parity you can sanely do raid6 on 100 drives, using
>> a logical stripe size that is much smaller than 100.
>> When recovering a single drive, the 10-groups-of-10 would put heavy load
>> on 9 other drives, while the decluster approach puts light load on 99
>> other drives.  No matter how clever md is at throttling recovery, I
>> would still rather distribute the load so that md has an easier job.
>
> Not offering to do it ... :-)
>
> But that sounds a bit like linux raid-10. Could a simple approach be to
> do something like "raid-6,11,100", ie raid-6 with 9 data chunks, two
> parity, striped across 100 drives? Okay, it's not as good as the
> decluster approach, but it would spread the stress of a rebuild across
> 20 drives, not 10. And probably be fairly easy to implement.

If you did that, I think you would be about 80% of the way to fully
declustered-parity RAID.
If you then tweak the math a bit so that one stripe would was

A1 A2 A3 A4 B1 B2 B3 B4 C1 C2 C3 C4 ....

and the next

A1 C1 A2 C2 A3 C3 A4 C4 B1 D1 B2 D2 ....

and then

A1 B1 C1 D1 A2 B2 C2 D2 A3 B3 C3 D3 ....

                  XX
                  
When Ax are a logical stripe and Bx are the next,  then you have a
slightly better distribution.  If device XX fails then the reads needed
for the first stripe mostly come from different drives than those for
the second stripe, which are mostly different again for the 3rd stripe.

Presumably the CRUSH algorithm (which I only skim-read once about a year
ago) formalizes how to do this, and does it better.
Once you have the data handling in place for your proposal, it should be
little more than replacing a couple of calculations to get the full
solution.

NeilBrown

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 832 bytes --]