Path-spec - fast circuits

[Mike Perry]

Thus spake ilter yüksel (ilteryuksel at gmail.com):

> >"No, you can only select 'Dual' flagged exit nodes (Exit+Guard) for the
> guard position.
> >Hence it doesn't make sense to have a Wge weight."
> 
> You said that "only" but, depends on your changes for dir-spec, it seems
> that we can select dual flagged, non-flagged and guard-flagged nodes for the
> guard position as below;
> 
> Wgg - Weight for Guard-flagged nodes in the guard position
> Wgm - Weight for non-flagged nodes in the guard Position
> Wgd - Weight for Guard+Exit-flagged nodes in the guard Position

Wgm exists for bridges, which don't have the Guard flag.

> Therefore i couldn't understand why there isn't Wge integer value on your
> changes. I think it seems possible that we can select exit-flagged nodes in
> guard position if guards are scarce. What do you think?

No, the proper way to handle this is to tune our algorithms that hand
out the Guard flag to ensure that Guards are always plentiful. These
algorithms are centralized and run on the directory authorities.

> On the otherhand did you simulate your changes? If yes, could you share
> results of the simulations?

Not yet. I have verified that the formulas produce balanced results,
and that my solutions do satisfy the system of equations. I plan on
doing selection tests once the system is deployed to verify that
clients are actually performing selection at the rates specified by
the weights.

Large scale simulations are hard to do. Ideally the academic groups
that work on Tor in various universities could pool their efforts and
maintain a common reference Tor Testing Network. It seems like every
researcher is always making their own small Tor network of various
sizes and load..

-- 
Mike Perry
Mad Computer Scientist
fscked.org evil labs
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL: <http://lists.torproject.org/pipermail/tor-talk/attachments/20100301/cb0fa30d/attachment.pgp>