I will try to fill in the gaps ...
On Mon, 2017-05-01 at 19:53 -0400, Steve Dougherty wrote: > > -------- Original Message -------- > > Subject: Re: Questions on Freenet - Second Try > > Local Time: May 1, 2017 7:43 PM > > UTC Time: May 1, 2017 11:43 PM > > From: st...@asksteved.com > > To: Discussion of development issues <devl@freenetproject.org>, phil > > ipp.spei...@uni-ulm.de <philipp.spei...@uni-ulm.de> > > > > > -------- Original Message -------- > > > Subject: Questions on Freenet - Second Try > > > Local Time: April 25, 2017 8:14 AM > > > UTC Time: April 25, 2017 12:14 PM > > > From: philipp.spei...@uni-ulm.de > > > To: Support <supp...@freenetproject.org> > > > Devl@freenetproject.org > > > Path Folding and Connection Managment: > > > > > > - Is it required that the data source actually starts to send a > > > "path > > > folding" offer back along the path or might any node on the path > > > simply decide to send its node reference back along the path, > > > hoping > > > to get new connections? > > > > It is my guess that a path folding offer can come from any node > > along the path. > > It is also how I remember it working. > > > - Does the data source always send the "path folding" offer back > > > or is > > > it possible to omit this. If the latter is the case, does this > > > imply > > > that there is no chance for creating new connections on this > > > particular path? > > > > I think it can be omitted. > > It can definitely be omitted (darknet), and as said above, any node on the path might make an offer. > > > - If one node answers the "path folding" offer and a new > > > connection is > > > created, is this "path folding" offer then passed back further > > > along > > > the path or does it stop at that node? > > > > I don't know. > > The request backtracks "the normal way" back to the initiator (taking a shortcut would cause havoc as it wouldn't kill the request). The offer made by the nodes on the paths are encrypted against the initiator's key IIRC (so only the initiator can decrypt/decide to peer). Several offers can be "bundled" with the response. It would be good to look into the details here... > > > - When connecting to a new node, how is the Friend-of-a-Friend > > > information (i.e. the locations of the node's neighbors) > > > exchanged? > > The only message type I've noticed that looks relevant is this one: ht > tps://github.com/freenet/fred/blob/build01478/src/freenet/io/comm/DMT. > java#L1345 > > I have not verified it's actually in use as such. > There is that and a few other things sent upon handshake https://github.com/freenet/fred/blob/next/src/freenet/node/PeerNode.java#L2258 > > > - Churn leads to frequent changes in the neighborhood of a node. > > > How > > > is the Friend-of-a-Friend information maintained to keep the > > > maintenance overhead small? > > You mean in terms of transfer of that information? > I don't think that we do any better than the obvious here. We send to each of our peers our location and a list of all of our peer's locations each time one of our connection status changes. I don't think that we do so when we are notified that one of our peer's peer location has changed (but check, don't take my word for it). (150+1)*8 bytes maximum I don't remember if we throttle it; I don't think we do; maybe we should. > > > Freenet's Load Balancing: > > > > > > - Relaying a request requires to keep track of it, i.e. the node > > > needs > > > to store information to be able to return the corresponding > > > response > > > at a later time. Due to resource limitations, a node cannot > > > accept > > > arbitrary requests and maintain state for them. > > > > > > -> So, requests can actually decide to reject a request depending > > > on > > > their current load. The corresponding connection is marked as > > > "backed > > > off", i.e. an exponential growing backoff time is determined to > > > pause > > > the connection. No requests are sent to the node during the > > > backoff > > > time to prevent putting further load on it. > > > -> However, Matthew Toseland writes in his work > > > http://www.toselandcs.co.uk/flogmirror/mjt92-diss-final.pdf that > > > also > > > a message is sent back to the request initiator in this case. > > > => I would like to know, what is the purpose of this message? > > https://github.com/freenet/fred/blob/build01478/src/freenet/io/comm/DM > T.java#L473 ? > > My expectation is to have an explicit feedback mechanism instead of an > implicit one based on lack of results. > s/requests/nodes/ on the question The purpose is to let the originator know that it has been backed off (and it should slow down). We can't prevent it from making further request but we can let it know that anything sent within the backoff period will be ignored. > > > => Does the request initiator react in some way in this situation? > > I'm not certain about that either. I would expect it prompts backoff; > there are probes for RejectedOverload percentages. > Yes, well-behaved originators will keep track of the rejected/backoff peers and temporarily exclude them from their routing tables. Florent
signature.asc
Description: This is a digitally signed message part
