Firstly, I'm not talking about completely separating the two networks. Please read the whole email before commenting on it.
Secondly, it is entirely reasonable to expect that a true darknet and an opennet will be two separate small world networks, which have completely different properties. Requests on the darknet would benefit from checking the darknet first, because data is often cached within the darknet (due to related interests, for example), and a darknet will often have very few links to the opennet, so there is a major performance penalty for the transition. What I propose is that requests on the attached darknet are routed first within it, and then forwarded out to the wider opennet. On the other hand, opennet requests which get routed to an attached darknet will often be falling down a black hole, for the same reason: That there is limited connectivity between the two networks. Our location swapping algorithm will work if it is running on a small world network. I doubt very much that it will work well if it is run on two small world networks which are not closely connected. Thirdly, yes it will break load limiting. It will definitely allow for sybil attacks (harvest, connect to as many nodes as possible, ignore all incoming requests), aimed not at disrupting the network but at achieving maximimal local performance (at the expense of others). We see this sort of trouble with BitTorrent clients. Token passing load limiting / balancing is not designed for opennet, and it's our best option at present. As Mrogers said: (Replying to me) > > The darknet is Sybil-resistant. However most people hear about freenet > > from slashdot (IMHO this is bad; we shouldn't be reliant on perpetual > > slashdottings, but that is unfortunately the current situation), so we > > need opennet. > > That doesn't follow - Slashdotters have friends too (OK, don't quote me > on that). Telephones caught on even though you needed to know someone > who had one before they became useful. Same with email and IM. If > Freenet supported instant messaging and shared folders I'd have a good > reason to invite my friends and family to start using it - they may not > be interested in anonymity but they're interested in privacy, and their > privacy can provide anonymity for others. Fourthly, the darknet *will* expand by itself, and some of it will be organic growth. There are a number of major factors limiting its growth right now: a) The network is small (feeds into b) b) There is little content (feeds into a) c) Inserting is slow (feeds into a) d) Insertion bugs (feeds into a) e) General bugs (feeds into a) f) Intrusive; no way to turn on and off easily (feeds into a) g) Security is uncertain (feeds into a, b) h) If a node is down for a while, it won't automatically update (feeds into a) Most of these are being worked on. Fifthly, various 0.8 security mechanisms (securing location swapping, premix routing) will not work easily/safely with opennet. Thus, I continue to take the view that we should not implement opennet before we have solved some of the current major problems, and got some real opennet simulations, and I further suggest that *maybe* it would be a good idea for opennet and darknet routing to be logically separate, while linked: darknet requests try the darknet first, inserts go to both. On Mon, Aug 14, 2006 at 08:34:53PM -0700, Ian Clarke wrote: > On 14 Aug 2006, at 13:17, Matthew Toseland wrote: > >Should darknet and opennet be semi-separate networks? > > I'm pretty sure that the answer to this question is "absolutely not". > > >IMHO we will need some sort of mechanism for dealing with the case > >where > >we have two networks, with limited connectivity between them. These > >would be "semi-separate" networks. This can occur with two > >darknets; the > >Chinese darknet and the Euroamerican darknet would probably have few > >links between them. But it may also be the case with darknet and > >opennet. > > Why? What is the advantage of separating the networks? As I see it > this would only reduce the utility of both networks. > > >The reason I suggest this for darknet vs opennet is this: > >- The optimal routing algorithms for darknet and opennet may not be > >the > > same; on darknet, location swapping is crucial, but on opennet, > >it is > > unnecessary. > > So what? Location swapping only occurs when it will improve the > network topology (except in rare circumstances), so if opennet has > already done a good job of achieving a desirable network topology, > then location swapping will leave it alone, it certainly shouldn't > hurt it. > > >- The load limiting algorithms proposed for darknet will need > > considerable adaptation for opennet. > > Why? OpenNet is simply about automating (and securing) a process > that is already occurring now, the difference is that now people are > using extremely inconvenient kludges like #freenet-refs, without > regard to achieving an appropriate network topology whereas opennet > will be extremely convenient for users, and should achieve an ideal > network topology. From most perspectives, including that of the load > limiting algorithm, opennet should make its life easier - it > certainly shouldn't be any worse than the kludged opennet we have today. > > >- There is no real reason to expect the two networks to make any > >sort of > > sense topologically when put together. > > Why not? We have two adaptive algorithms both trying to achieve the > same goal - a log(d) network topology. Both of them have the quality > that their effect on the topology is proportional to the difference > between the current topology and its desired state. Because of this, > the more I think about it, the more convinced I am that both > algorithms will complement each other, rather than step on each > other's toes. > > Separating the opennet from the darknet will reduce the utility of > both networks, I suspect it will probably result in almost nobody > using the darknet at all, I assume that isn't what you are trying to > achieve? > > I think its time we acknowledged that until we provide the option of > proper opennet functionality, we won't actually have a darknet, since > until we provide a proper solution, everyone is just kludging their > own extremely cumbersome opennet using #freenet-refs, and other tools > (none of which are likely to lead to a desired network topology). > > Ian. > > Ian Clarke: Co-Founder & Chief Scientist Revver, Inc. > phone: 323.871.2828 | personal blog - http://locut.us/blog > > _______________________________________________ > Devl mailing list > Devl@freenetproject.org > http://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl -- Matthew J Toseland - [EMAIL PROTECTED] Freenet Project Official Codemonkey - http://freenetproject.org/ ICTHUS - Nothing is impossible. Our Boss says so.
signature.asc
Description: Digital signature
_______________________________________________ Devl mailing list Devl@freenetproject.org http://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl
