[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: $subject



David Woodhouse wrote:

> On Sat, 3 Mar 2001, Vipin Malik wrote:
>
> > >Compare that with the probability that the sky will fall on your head
> > >anyway. Do we still care?
> >
> > If I understand you correctly, you are arguing the case that reading
> > the same "questionable" sector twice is good enough even if you are
> > paranoid, and that we will catch at least 1 flipped bit, and the
> > probability of missing that is so low that we probably don't care for
> > that case.
>
> Basically, yes. It's _vaguely_ feasible that the sector was 99.9999%
> through the erase, and there are only one or two bits which only have
> "half an electron". In that case, perhaps we wouldn't catch them, and we'd
> later end up with single-bit errors in a couple of nodes, which we ought
> to be able to deal with _anyway_. We should probably be reading back nodes
> after write and checking the data were written properly anyway.

I implemented the read and compare twice and guess what? Power cycle #219
managed to give my read compare twice the slip and kernel panicked my kernel
due to "flipping bits" that were not caught in a sector :)

I guess that we don't understand the mechanism of "flipping" and that the
probability of missing even 1 flipping bit in an entire sector is not that
low!
(Heck 1 cycle out of 219 practically speaking :)

I've implemented a NUM_REREADS based method now, with NUM_REREADS
set to 4 at this time. Let's see how far this goes.

Vipin


To unsubscribe from this list: send the line "unsubscribe jffs-dev" in
the body of a message to majordomo@xxxxxxx.com