Re: [Mldonkey-users] feature suggestion: forward error correction

[Goswin Brederlow]

Ignacy Gawedzki <address@hidden> writes:

> On Thu, Sep 05, 2002 at 02:45:45AM +0200, thus spake Christian Brandt:
> There need not necessarily be some complete copy available on some host
> in order to avoid the problem. In fact the union of all the chunks
> available must be the complete file.
> 
> Then we must rather concentrate on how to spread the file so that this
> condition holds as soon as possible. This could be achieved by letting
> the uploader decide what chunk to send to a downloader based on a
> latter's list of missing chunks. I must confess that I haven't looked at
> the protocol yet, so I dunno how it is actually done. An alternative
> solution seems to be that the downloader picks the chunks to ask for
> more randomly.

As far as they descibe it for edonkey the client gets or askes for a
random chunk but priotises on the first and last.

mldonkey priotises on the last and then gets the file from start to
end.


The optimum chunk to send is probably the chunk that the least number
of clients have. With FEC this can be maximised by sending a chunk
noone else has, one freshly generated (if one has the complete file).

> In this way, the availability of chunks should be uniform and there
> should not be any "rare" chunks anymore.

I was thinking about only starting an upload if a file was seen
complete at some time. The client would keep track of all chunks it
sees and once it saw them all it actually starts downloading. Not all
chunks have to be available right now, having seem them at some time
should be enough.

That should stop the spread of files that never finish due to a
missing chunk.

MfG
        Goswin