Re: Location churn: empirical data
Jano <[email protected]>
| Newsgroups | gmane.network.freenet.technical |
|---|---|
| Message-ID | <[email protected]> |
Matthew Toseland wrote: > On Thursday 15 May 2008 11:46, Jano wrote: >> Daniel Cheng wrote: >> >> > 2008/5/14 Jano <[email protected]>: >> >> Matthew Toseland wrote: >> >> >> >>> Below are links to my node's location logs. A lot of the location churn > is >> >>> within roughly a slice of 10% of the keyspace, and the central signal is >> >>> reasonably clean... but this degree of variation is going to cause > problems >> >>> with routing and especially data retention IMHO (on a network of several >> >>> thousand nodes, with *at least* 1000 nodes online at a time). There were > a >> >>> whole bunch of resets during the week 19 April to 26 April, and on the >> >>> 24th, RC2 was released; on May 8th, 0.7.0 was released. These may > explain >> >>> some but not all of the long jumps... >> >>> >> >>> http://amphibian.dyndns.org/location.log.txt >> >>> http://amphibian.dyndns.org/locations.png >> >> >> >> See mine attached. This node has not been steadly 24/7 up as of late, > which >> >> can be more typical. It bounces a bit... >> >> >> > >> > How many opennet connection do you have? >> >> This node is pure opennet. So 20, I think. > > Ok, in that case each time there is long downtime it will have to reseed... > even so, seeding is directional... but your new nodes won't be the same as > your old nodes, so it may partly explain it. I thought that even in opennet, the node would try to reuse its old peers after restarting. Is this so? Still, I understand that long downtimes may cause more of your peers to vanish.