logo       

Re: Disk cache: msg#00028

network.bit-torrent.libtorrent

Subject: Re: Disk cache

> > Ah, I see... I had not thought of such a scenario. Very interesting.
> >
> > So, in a torrent like you described (1 seed, 1 hidden/offline peer, 1
> > blind peer), assuming downloaders get 1/2 of the file each, they would
> > be able to finish only if they have complementary halves of the file.
> >
> > With a randomization at the piece level, with say, 1,000 pieces in
> > total, each of the downloaders need the right 500 pieces to make the
> > torrent healthy.
> >
> > The probability for the blind client downloading the right 500 pieces,
> > not knowing which 500 the hidden peer has, are about 1/(2^1000), since
> > each piece has 1/2 chance of being the right piece, and that there are
> > 1000 guesses to be made.
> >
> > I still think sequential downloading gives the torrent a better chance
> > for healing:
> >
> > Say that each client downloads quarters of the file in random
> > order. The
> > pieces in each quarter are downloaded sequentially. (ie, all the
> > pieces
> > in quarter 1 are downloaded in sequential order, but the file may be
> > downloaded as Q2,Q4,Q3,Q1)
> >
> > In this case, if the torrent behaves like you described above, the
> > chances for the torrent remaining healthy with the blind and hidden
> > peers present, but seed absent, are about 1/(2^4), since there are 4
> > guesses to be made. If each of the clients has 1/2 of the file when
> > the
> > seed goes away, they have a 1/16 chance of finishing.
> >
> > So comparing a download strategy of piece-by-piece-randomly (PBPR) or
> > quarter-by-quarter-randomly (QBQR), the QBQR strategy has a 6.25%
> > chance
> > of finishing, while PBPR has no (1/(2^1000)) chance of finishing.
> >
> > I think even the 6.25% probability can be improved:
> >
> > If the file is downloaded in halves instead, there is a 25% chance of
> > finishing this download. Perhaps each client can decide whether to
> > download halves, quarters, eights or whatever, depending on how many
> > peers it encounters:
> >
> > 1 seed, 0 peers = download halves
> > 1 seed, 1 peer = download quarters
> > 2 seeds, 2 peers = download eights
> >
> > What do you think?
>
There is superseeding ... which defeats this appoarch totally.

MassaRoddel


-------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642


<Prev in Thread] Current Thread [Next in Thread>
Google Custom Search

News | FAQ | advertise