On Wed, Sep 2, 2009 at 6:29 PM, Matthew Toseland<toad at amphibian.dyndns.org> wrote: > On Wednesday 02 September 2009 20:21:02 Michael Yip wrote: >> Hi Matt, >> >> Attached is my locations file. >> >> I've been running an experiment of a small Darknet of 4 nodes and my >> node has 3 peers. They were not fully interconnected. >> >> A peer of mine sends 1000 requests and one of the 3 nodes (including >> mine, randomly chosen) was restarted after every 10 requests have been >> sent by my peer. >> >> The frequency of the location change tells me that location of my node >> changes pretty frequently. Also, the location can vary by a huge amount e.g. >> >> 02-Sep-2009 00:50:44 : 0.8453261772531677 >> 02-Sep-2009 00:51:19 : 0.006713131335581379 >> >> This would limit the ability for a node to specialize it's datastore right? > > Swapping will only stabilise if the network is actually small-world. A > network of 4 nodes is unlikely to meet the criterion. :|
Also, those locations only differ by about 0.26. On a 4-node network, it's hard to argue that the node's location changed very much. I suggest that if you want meaningful simulation results, you should probably test on a network where number of nodes > (average node degree)^x, for some x > 1.5, and probably > 2. Similarly, I would test with diameter at least 3, preferably 4 or higher. Also, the fraction of nodes with degree < 3 should be very small, and I recommend an average degree of at least 4. You'll also need to ensure that the network exhibits the small world property. If testing on a particularly tiny network, I also recommend you consider carefully what the appropriate htl value is (it should probably be smaller than on the regular Freenet), and whether you want to disable probabilistic htl. Simulation data on location churn in the face of uptime < 100%, and its impact on data retrieval, would be nice to see. Evan Daniel >> >> Thanks a lot, >> >> Michael >> >> >> >> Matthew Toseland wrote: >> > On Wednesday 02 September 2009 18:40:21 Michael Yip wrote: >> > >> >> So that means that if one of my peers restarts, location swapping would >> >> occur and as a result, the location of my node may change to avoid any >> >> collision? >> >> >> > >> > No, not usually. >> > >> >> >> >> Matthew Toseland wrote: >> >> >> >>> On Wednesday 02 September 2009 15:57:09 Michael Yip wrote: >> >>> >> >>> >> >>>> Hi guys, >> >>>> >> >>>> Can someone explain to me what is logged in the file locations.log.txt? >> >>>> >> >>>> Here's my guess. It logs the location changes of my node and this >> >>>> prevents my node from specializing in storing a subset of the keyspace? >> >>>> I'm guessing this only happens in Darknet as it would decrease >> >>>> performance... >> >>>> >> >>>> Am I correct? >> >>>> >> >>>> >> >>> On darknet nodes, the location can change due to location swapping. This >> >>> is necessary to make routing work on a darknet. I suggest you look at >> >>> some of the stuff in the papers section of the website, starting with >> >>> the CCC video: >> >>> >> >>> http://freenetproject.org/papers.html >> >>> http://freenetproject.org/22c3vid.html >> >>> >> >>> >> >>>> Thanks a lot, >> >>>> >> >>>> Michael > > _______________________________________________ > Devl mailing list > Devl at freenetproject.org > http://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl >
