new class / geom_raid5

[R. B. Riddick]

--- Stanislav Sedov <stas at 310.ru> wrote:
> îÁ Thu, 6 Jul 2006 15:42:09 -0700 (PDT)
> "R. B. Riddick" <arne_woerner at yahoo.com> ÚÁÐÉÓÁÎÏ:
> > My proposal for
> >   geom_raid5
> > is ready so far...
> > 
> > If somebody wants to test it, he or she can download it from
> >   http://home.tiscali.de./cmdr_faako/geom_raid5.tbz
> 
> How do you handle data recovery in case of crash of one of providers?
>
Hmm... As far as I understood the RAID5 idea, it is the following "trick":
A XOR B XOR C = P
A = P XOR B XOR C
B = P XOR A XOR C
C = P XOR A XOR B

> All raid5 implemetations use different ECC algorithms for this,
> like Reed-Solomon codes etc.
>
Hmm... I dont know so much about coding theory... I know that ECC means "error
correction code" and I think the above formula describes an ECC, that can
correct a missing bit, if we already know which bit is missing...

E. g.:
If we have the valid codewords 10 and 01, then we cannot say what 11 means,
unless somebody tells us the second bit is a really very bad bit.

> I haven't looked deeply into the code, but it seems there is no
> any such algrithms implemeted there. Do you just duplicate data
> on all providers?
> 
Hmm... "duplicate data on all providers"... That sounds like a RAID1 /
MIRROR...

I am sorry, that my code is so bad, that it looks like MIRROR code...

Maybe you should have a look at my newest version (I changed some things, but
it is still mostly untested)?

I have heard, that code inspection and tests should be done by somebody, who
did not write it... So I would be glad, if somebody helps me here (of course I
will do some tests, too, although I am a little bit annoyed by the BIOS-beeps
already)...

Should I use more but shorter functions?

-Arne


__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com