On Tue, Apr 11, 2006 at 08:35:23PM +0100, Michael Rogers wrote: > -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA1 > > Matthew Toseland wrote: > > Well, the advantage of what we have is that it is based on well known, > > well studied algorithms which are known to work in practice. So any > > replacement would have to either be some means of enforcing the existing > > algorithms (which would require being able to identify local requests, > > which is bad... unless they are premix routed before becoming local > > requests), or something that had been *extensively* simulated and proven > > to work. > > There's a lot of work on fair queueing that you could look at... but I > mean a lot. :-/ It's not my area but the summary here looks useful: > > http://stelvio.univ.trieste.it/~vitez/Queuing.htm
Nice. Of course every time we read a paper we risk violating a patent. :( Statistically speaking I expect anything published since 1986 will be patented and therefore off limits. (This is a *GREAT* way for the establishment to oppress free software without really trying). Anyway queueing is only a small part of the problem. The big issue is how to propagate load back to the original sender of the request, or at least to approximate this, while either detecting or preventing flooding by dishonest peers, and ideally without giving away which node is the original requestor (but that may be a contradiction). Our current algorithm propagates load back to the original requestor, however it does not prevent flooding and gives away the original requestor to his peers. Well-studied as it is used in TCP. MRIs (Minimum Request Interval; an explicit statement of how fast nodes can send requests, implemented in 0.5, details on request) don't give away the requestor, and do prevent flooding (if properly implemented to enable enforcement), however they don't adequately propagate load back through the network (beyond one hop). At least they don't if we can't make the MRI of a node dependant on the MRI's of its peers. Any other possibilities? Anything we'd use must be robust against both attacks and normal traffic, it shouldn't be overly complex, and bear in mind that load balancing/limiting has been *the* freenet bugbear for years, along with routing (and the interaction between the two). > > Cheers, > Michael -- Matthew J Toseland - toad at amphibian.dyndns.org Freenet Project Official Codemonkey - http://freenetproject.org/ ICTHUS - Nothing is impossible. Our Boss says so. -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 189 bytes Desc: Digital signature URL: <https://emu.freenetproject.org/pipermail/devl/attachments/20060411/13c9ffcc/attachment.pgp>
