[Cryptography] Cubbit

Nemo nemo at self-evident.org
Mon Jun 8 14:09:28 EDT 2020

On Sun, Jun 7, 2020 at 3:28 PM Tony Patti <crypto at glassblower.info> wrote:
> > Jerry Leichter <leichter at lrw.com> writes:
> >
> > > The description Cubbit provides says that they distribute 36 copies in
> > > a Reed-Solomon code that allows recovery from any 24 copies.  But one
> > > way or another ... 36 copies requires that, somewhere, there be 35
> > > times the space of the original copy to provide the redundancy.
> >
> >Nemo <nemo at self-evident.org> writes:
> >
> > We have a RAID6 at work that uses 12 1TB drives. It provides a full 10TB
> of usable storage, and any two of those 12 drives can fail without causing
> us any data loss whatsoever.
> >
> > How does this fit with your analysis?
> >
> >  - Nemo
> RAID5 and especially RAID6 have TERRIBLE write performance, which you can
> quantify here: https://wintelguy.com/raidperf.pl

RAIDx write performance is fine if you work with files in the hundreds of
gigabytes and write to them in large sequential chunks, which we do. I will
stop here before I dox myself.

The point is that 2x, 3x, 4x, etc. redundancy is quite possible without
doubling, tripling, or quadrupling your storage, contra Jerry's analysis. I
have not studied Cubbit, but it sounds like you can lose a third of your 36
"partners" and still recover all of your data at a storage increase of 1.5x.

 - Nemo
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://www.metzdowd.com/pipermail/cryptography/attachments/20200608/86a8a6b7/attachment.htm>

More information about the cryptography mailing list