On Sat, Jun 07, 2003 at 10:42:26AM +0200, Thomas Leske wrote: > fish wrote: > > On Fri, Jun 06, 2003 at 07:57:02PM +0200, Thomas Leske wrote: > > > One could also solve the two problems without containers: > > > 1) completeness: > > > Make FEC work for sets of small files. The regular files are > > > just inserted normally. The zip archiv is only generated on the side > of > > > the inserter or reader for creating respectively using the > > > error correcting blocks but the archiv is not stored on freenet. > > > > I should point out, that if your container is large enough to FEC (that's > > around a meg), then your container is too fucking big - in ideal > conditions, > > (i.e. not freenet), a 56k modem (which are very common, despite what some > of > > the broadband elite think) will take around 5-10 minutes to fetch your > data. > > But this is a shortcoming of the current FEC implementation that was > designed > to protect large splitfiles. If you choose a smaller block size in the > fec algorithm, it could be used on smaller files as well. There is a quite > flexible tool that could do the job independent of fproxy: > http://parchive.sourceforge.net/ > But an in-freenet solution would be much more elegant, because the site can > be browsed via fproxy and you can reinsert missing chunks. > Of course you would not base the FEC Encoding on the zip archive format, but > extend the current FEC meta data to allow different sizes for each chunk and > an arbitrary block size. The chunks are concatenated to a large file that is > split into blocks in order to determine the correcting blocks. It may also > be > useful, to allow zero padding after a chunk up to the next block boundery. > If the block size is quite small, you would combine the correcting blocks to > a number of larger chunks. The second version of parchive allows correcting > chunks of increasing sizes of 2^n. Thus you have to download just as much > error correction information as you need. > > > And that[s of course on the low end of the FEC scale. Files this large > > should not be allowed to be containered. > > Note that there is no container in my proposal that everyone has to > download. > One can also just browse a site in the usual and unreliable way. Only if you > what to fetch the whole site or you really need a certain missing file > you use the FEC feature (and combine the chunks to a container in your > client).
The result of this is that all the check blocks will fall out of freenet because of not being used. This is why we always download a random m out of n blocks in the splitfile fetcher. > > Thomas -- Matthew J Toseland - [EMAIL PROTECTED] Freenet Project Official Codemonkey - http://freenetproject.org/ GPG key lost in last few weeks, new key on keyservers ICTHUS - Nothing is impossible in him whom we trust
pgp00000.pgp
Description: PGP signature