Very interesting - keep up the good work! Ian.
On Sat, May 5, 2012 at 12:43 AM, Steve Dougherty <steve at asksteved.com>wrote: > -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA1 > > I've completed an initial run of simulation work on probes. The code is > available, [1] as well as the simulation results from which the plots > were generated. [2] The point of immediate interest though is the plots > themselves, [3] which show the predicted network coverage of different > probe routing techniques on networks with ideal degree distribution > (more on this later) and that following the degree distribution of > Freenet as measured. [4] Link lengths and locations do not factor into > this simulation because probes take only degree into account and are not > seeking any given destination; their goal is only to average out to > distribute endpoints uniformly throughout the network. > > The ideal network distribution has each node add a fixed number of > remote connections without regard for the number of connections it or > the nodes it's connecting to have. I don't know whether this or having > each node have the same number of total connections is the ideal. The > results of the simulation did not appear to greatly change with > network size, as shown by the consistent behavior between the 12,000 > and 45,000 node versions of the MH-corrected degree-conforming > simulation. [5] > > As expected, the plots suggest that using Metropolis-Hastings correction > will be an immense improvement in endpoint uniformity over the current > uniform random routing, but specifically suggest that an HTL of around > 20 hops is close enough to a baseline uniform endpoint probability to be > a good starting point. I've noticed that these CDFs aren't a very good > format for demonstrating closeness of distributions, given overlapping > lines, but I don't understand the Kolmogorov?Smirnov test yet, so I'm > planning to just use these results as a guideline and begin implementing > the new probes next week. > > The gnuplot scripts to generate the degree and link length distribution > plots are part of pyProbe, [6] and GNU parallel [7] is used in test.sh > to run simulations in parallel. In the simulator source there are > scripts from an earlier effort to plot coverage as percentages, but that > was even less clear than the CDFs. > > Comments and suggestions are very welcome! > > Thanks, > operhiem1 > > [1] https://github.com/Thynix/routing-simulator/tree/dev > [2] http://asksteved.com/plot-source.tar.xz > [3] http://imgur.com/a/Z8SBS#2 > [4] http://i.imgur.com/ehfBP.png > [5] http://i.imgur.com/rtRIB.png > [6] https://github.com/Thynix/pyProbe > [7] http://www.gnu.org/software/parallel/ > -----BEGIN PGP SIGNATURE----- > Version: GnuPG v1.4.11 (GNU/Linux) > > iQIcBAEBAgAGBQJPpL3dAAoJECLJP19KqmFuNn0QAMsA4nzk6AfPf8pIqrmoEW8U > 2jcc7L3KnUkCIgvh9FyhJkZ9Fm42zCoqgxXmyavM9T18ZO52eYaNMaSfkA5FWltk > iBElymF7ZCGd3ERX9XPirbXGDeMbpNsFbVHFoJbqKzb94MrnSUivLsVQz0Nl1KOJ > g1yfYdA4RK3ywYIvwS7nWkIIrxhuik/Jzjaq5cuqY2L6i3DgiM9gjYweyJLpzt6r > k/mRNOuKTI0MSdqMWclBFXOEOzTg/vZKZSvvslpZRwt0Opp+nK9VKBMVzvqiqUpr > G9EEke4vPqU8OdWffxqu3nF5ZXlr4aB3mWw6B7zimE+7C3Wvk3oQHxxv/p/PqD96 > GQ/sUbkFERSv/SnMDCuz8BVoPNihTyohvRJmeW92P2KpFCJ7Ynsx1uC6XLKDQVIO > Qxds7EUKkdEQaEbNYRKMkzx9qzOszRZlcvLElX2Fgw15KvTMKmMDb/7t1DpbBysY > tl7JnkYW6crq3nvBpWu3JFmSOYERhEzzKxkRsE76DVzkBz35AYOb1ZTLx06mEgP4 > F8HFs31Ra8LNlVCoN5jEHW3WhUIVkVtx8zauXGOtjJuY4ePhEXS9TvXOKAbvxMiA > d/Nu78MORKBdq1repSMIcCLUl1Ya0AT0BEugvJ4KyKPScl0JL0GPOiFBG8Dr01GZ > pQtYR4VDpcLlPzPkq1xj > =apcO > -----END PGP SIGNATURE----- > _______________________________________________ > Devl mailing list > Devl at freenetproject.org > https://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl > -- Ian Clarke Personal blog: http://blog.locut.us/ -------------- next part -------------- An HTML attachment was scrubbed... URL: <https://emu.freenetproject.org/pipermail/devl/attachments/20120510/e9aa10f6/attachment.html>
