On Thu, Apr 13, 2006 at 04:45:45PM +0100, Michael Rogers wrote: > -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA1 >
Your first diagram:
E F
| |
A---B---C---D
> > That seems a bit much. Can it be avoided? I worry that we won't get
> > enough data if we have to keep one for each pair.
>
> The rate's only adjusted when we get data so I don't think it will be a
> problem - if there's very little traffic between a given pair of
> neighbours then the rate isn't so important anyway.
I mean we won't get enough information to have a useful and accurate
rate for each pair.
>
> > How much of a problem is the above?
>
> I'm not sure how much of a problem it is - the link BC will be throttled
> in response to congestion along ABCD, but not as much as the links AB or
> CD, because successful requests from E to F will offset the reduction
> caused by unsuccessful requests from A to D.
This is not a serious problem, because we are not really dealing with a
point to point routed network here. The same proportion of requests from E
and from A will reach F. If D is overloaded, then everyone needs to slow
down, because *everyone* is talking to D. If they are further away they
probably talk to D less and so need to slow down less.
>
> In general I suppose the more paths there are sharing a link, the more
> the reductions caused by one overloaded path will be offset by increases
> caused by other paths. This means the other paths will be less affected,
> but it could also mean the throttling won't propagate back to the sender
> - - if the link BC doesn't slow down much then the link AB won't slow down
> much either...
As I said it's not really about paths, at least it isn't at present. We
may need to deal with path-based load balancing in 0.8, but in 0.7 we
don't need to worry too much about it because requests go everywhere
more or less at random. Admittedly if we start establishing long-lived
paths for anonymity reasons this could be more of a problem... But would
the algorithm work with the current usage of the network?
So, is there a problem with either this scheme or the originally
proposed one?
>
> >> F G
> >> | |
> >>A---B---C---D---E
> >
> >
> > What's the problem here?
>
> Even if C keeps a separate rate for each pair of neighbours, the paths
> ABCDE and FBCDG have the same previous and next hops at C, so if C
> reduces its rate from B to D in response to congestion at E then FBCDG
> will also be throttled.
IMHO this is not a problem at present.
>
> > My basic concern is whether it will get into freeze-down troubles with
> > loops (any real network will have many many loops).
>
> As long as loops in a given path can be detected, are loops in the
> topology a problem?
No. What might be a problem:
A-B-C-D-E-A (benzene ring)
Congestion at E, so E-D slows down, C-D slows down, B-C slows down, A-B
slows down, E-A slows down... This is only a problem if we overreact...
Would we?
>
> Cheers,
> Michael
--
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
