On 16 Oct 99, at 7:35, Jukka Santala wrote:
> [ ... snip ... ]
> Except that PrimeNet doesn't control the prize. This is the error everybody is doing.
> EFF is adminstrating the competition and prize, given by anonymous donaters to
> advance distributed computing / mathemathical algorithms on computers. PrimeNet is
> just one of the organizations (With largest changes known!) to get that prize, but it
> doesn't decide upon who gets it. The first person to present a prime filling the
> requirements will - and that's why result-files "few iterations short" will be worth
> more than their weight in gold.
I think, if someone tried to pull this particular stunt, PrimeNet
could inform EFF of the claimant's theft of the right to the
discovery (which is what it amounts to). Another unfortunate example
of the way in which large ca$h prizes can lead to unpleasantries, I'm
afraid.
>
> You said... about making the repository write-only. And you seem to completely miss
> the point; the discussion here has centered around how to solve the problem that
> PrimeNet doesn't have the bandwidth or storage space to backup the intermediary
> files! So my suggestion was "distributed storage". This will solve the backup
> problem, but yes the above-mentioned prize-claim and double-check problem remains.
> Unfortunately, I think in that respect we're stuck.
Well, I sort of _did_ miss that particular point. But I have a
solution. If you have a distributed filestore, controlled by a number
of different people, then you encrypt & hack up the file you're
saving and scatter pieces amongst the remote filestores. If you (the
assignment owner) use a method based on a key otherwise known only to
PrimeNet to scramble & scatter the file, the small chunk of data
stored on any other individual's system is useless to them for the
purposes of hijacking your work.
I think PrimeNet should know users' keys so that data can be
recovered in the event of the user defaulting or accidentally
destroying their own key (so that work done isn't lost). Presumably
we trust the people running PrimeNet!
An obvious extension to this idea is to use a RAID type technique
e.g. if you store each bit of a byte on a seperate host, so that the
file is scattered over 8 hosts, you could send the parity bit to a
ninth host. Then the distributed filestore is immune to loss of one
of its hosts (by reason of network failure or unsecured loss of its
filestore).
As for bandwidth and total filestore space needed - the first is, and
will probably remain for some time, a problem, at least for most home
users. 10,000 x 8 Mbyte save files is a lot of data by today's
standards; nevertheless, 4 20GB drives at $200 each would just about
cover it. (A random dip into a UK magazine shows Samsung 20.3GB
UDMA66 IDE at �144 (sterling) each, so US$10 per GB should be
achievable)
Regards
Brian Beesley
_________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
