Reg boffins: Help us answer this Big Blue RAID data recovery poser

IBM's research department haz released a research paper on RAID 5 that has intrigued and baffled our correspondent in equal measure.

The paper, by IBM Almaden researcher Mario Blaum, professes to solve a problem where RAID 5 is insufficient to recover data when two disks fail. Blaum's solution claims to do so better than a RAID 6 scheme. The paper's abstract says:

A construction of Partial Maximum Distance Separable (PMDS) and Sector-Disk (SD) codes extending RAID 5 with two extra parities is given, solving an open problem. Previous constructions relied on computer searches, while our constructions provide a theoretical solution to the problem.

Alerted at once to an insufficiency of cerebral expertise, your hack pressed on to the introduction. Its first sentence reads:

Consider an m×n array whose entries are elements in a finite field GF(2^b) [4] (in general, we could consider a field GF(p^b), p a prime number, but for simplicity, we constrain ourselves to binary fields).

Enough already! We can't determine if this IBM Research paper is hot stuff or an interesting mathematical side line with no real-world applicability.

So, having publicly admitted to our cerebral insufficiency, let's try crowd-sourcing (via you, dear Reg readers) to get an understanding of it.

The deal is we make the paper available here (PDF), you read it, and those of you still sentient and standing after doing that can post a plain language description of it in this Reg forum thread. Then everyone can benefit from your expertise, you become known as a …. wait for it … "Thought Leader" and get looked up to admiringly by your fellow commentards.

Also, I get to understand a bit better what Big Blue's maths boffin is on about and won't make a complete dick of myself writing about it. ®