Re: [Gzz-commits] manuscripts/storm article.rst

[Benja Fallenstein]

address@hidden wrote:
> Quoting Benja Fallenstein <address@hidden>:
>
>
> > Hi Hermanni,
> >
> > address@hidden wrote:
> >
> > > We have to be careful with distrubution of shares, espcially if used for
> >
> > better
> >
> > > data availability. You should read IPTPS '02 publication how distribution
> >
> > of
> >
> > > shares failed in Mojo Nation:
> > >
> > > http://zooko.com/IPTPS02.pdf
> > >
> > > They estimated the 'half-life' phenomenon wrong among participating
> >
> > peers.
> >
> > > However, they give a work around proposal (used in Kademlia) on this issue
> >
> > in text.
> >
> > I'm not planning to do a share-based system. I was just thinking that 
> > the *possibility* of such a system might be another reason for tree 
> > hashing. But downloading from multiple sources is the main one, of course.
>
>
>
> Ok. 
>
> Not directly related to this: I think CFS' downloads won't work in 'real-life'
> very well, since the mini blocks are shared among participants. As seen in Mojo
> Nation, half-life of the system is very small ;).

Hm, I *think* that CFS is more geared towards a network of 'servers'-- 
i.e., computers that are permanently connected to the Internet. 
OceanStore certainly is designed like this. But for 'real peers' I do 
think that they're better used only for sharing data they already store 
and storing items in a DHT (being replaced by other replicas when they 
go offline).

-b.