On Mon, Jun 16, 2003 at 08:46:06PM +0200, [EMAIL PROTECTED] wrote: > and a copy for devl@ > > >> don't over-specialize freenet, the disjunction file<->splitfile is ugly > >> enough [[[entropy's approach is imho better]]] > > > >Hey, do you mind posting a description of what entropy does differently. I > >have so far been unable to find ANYTHING in their docs that seems to > >distinguish them from Freenet. > > entropy splits up every file into 4k blocks and uses an algorithm comparable to > freenet's fec to create additional redundant blocks (fixed 50% redundancy > IIRC). > then all those 4k blocks and checkblocks are inserted into the net, just like we do > in freenet. > the difference is that every file >4k is fec-ed and thus small files, such as > html-pages and graphics are downloadable in parallel by requesting different blocks > and that those smaller files are more protected from falling of the net due > redundant blocks and reconstruction. > it uses a fixed block size of 4k, which may be too small to be effective, but the > main advantage is, that ALL blocks are 4k and thus are easier to handle and > the meta-algorithms such as routing and timeouts are more prdictive than using the > various block sizes from freenet > also i think their fec-algorithm (which experienced a rewrite some time ago) is > faster than the one used by freenet. entropy is coded in C(pp?) and so it might
It is however crap. It does not provide the sort of redundancy we get with Onion FEC from what I have heard. Also I am concerned that they do the splitfiles on too low a level, so it is visible to nodes, which is a bad thing. > be able to reuse their fec-code for freenet by reimplementing it with java and/or > compiling it to native for various platforms for additional speedup. (freenet's > fec also seems to be somewhat buggy, limited (128MB segments for files too lange) > and not always threadsafe?) You can scale it down with smaller blocks. We do this in the streaming over freenet stuff. You can scale it up with bigger blocks or with segments. It is not in itself noticeably buggy AFAIK (except the native FEC, which obviously would be faster if it worked). > > unfortumately entropy is merely a single-person project, so the algorithms used are > not tested very well and might also be proprietary. The algorithms used are according to what I heard not very good. Freenet inserts 1.5N files, which go to completely different routes and nodes. Any N of them can reconstruct the original file and all the check blocks. Entropy's "FEC" can only correct the loss of a few blocks. > > entropy's board can be found at "http://f27.parsimony.net/forum66166/index.htm"; > > please read the author's original documentation, because everything i wrote here is > from memory only! > > HTH -- 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. Our Boss says so.
pgp00000.pgp
Description: PGP signature
