How RAID-6 dual parity calculation works (opens in new tab)

(igoro.com)

55 pointsc0ff11y ago7 comments

7 comments

Here's someone doing the math to generate matrices for higher amounts of parity: http://www.spinics.net/lists/raid/msg44948.html

signa1111y ago

i would rather go with this : https://www.kernel.org/pub/linux/kernel/people/hpa/raid6.pdf for much better in depth coverage...

level11y ago

> This article is for computer engineers who would like to have a high-level understanding of how the dual parity calculation works, without diving into all of the mathematical details.

That is more in depth, but is a lot more math heavy, which the author specifically states they are trying to avoid.

shanemhansen11y ago

I come from a different math background, and it always surprises me people don't explain erasure encodings in terms of polynomials.

We all know that a polynomial of degree n is basically defined by n points. So a line is defined by 2 points.

If I need to encode the 2 pieces of info [a,b] I can just generate a polynomial ax+b and sample it at 2 points to recover the info. If I sample it at 3 points I can lose any one point and still recover [a,b]. The concept generalizes to higher dimensions and discrete values.

afaik that's basically it.

dmiladinov11y ago

From the article:

> This article is for computer engineers who would like to have a high-level understanding of how the dual parity calculation works, without diving into all of the mathematical details.

I believe that is the value proposition for reading: gaining an understanding of RAID-6 parity calculation while glossing over the math-y bits.

oofabz11y ago

Why does it need separate parity formulas? It would be easier to put the same simple xor in both parity blocks.

aYiZtr11y ago

Suppose you have four disks and lose two. Consider a set of blocks where you lost two data blocks and still have two parity blocks. The parity blocks would be identical and you would not be able to recover data for either of the two missing data blocks.

j / k navigate · click thread line to collapse

7 comments

sp33211y ago

Here's someone doing the math to generate matrices for higher amounts of parity: http://www.spinics.net/lists/raid/msg44948.html

signa1111y ago

i would rather go with this : https://www.kernel.org/pub/linux/kernel/people/hpa/raid6.pdf for much better in depth coverage...

level11y ago

> This article is for computer engineers who would like to have a high-level understanding of how the dual parity calculation works, without diving into all of the mathematical details.

That is more in depth, but is a lot more math heavy, which the author specifically states they are trying to avoid.

shanemhansen11y ago

I come from a different math background, and it always surprises me people don't explain erasure encodings in terms of polynomials.

We all know that a polynomial of degree n is basically defined by n points. So a line is defined by 2 points.

afaik that's basically it.

dmiladinov11y ago

From the article:

> This article is for computer engineers who would like to have a high-level understanding of how the dual parity calculation works, without diving into all of the mathematical details.

I believe that is the value proposition for reading: gaining an understanding of RAID-6 parity calculation while glossing over the math-y bits.

oofabz11y ago

Why does it need separate parity formulas? It would be easier to put the same simple xor in both parity blocks.

aYiZtr11y ago

j / k navigate · click thread line to collapse