[Author Prev][Author Next][Thread Prev][Thread Next][Author Index][Thread Index]
[tor-bugs] Re: #1294 [Tor-Tor server]: Bandwidth weights absent when D=0
#1294: Bandwidth weights absent when D=0
-----------------------------+----------------------------------------------
Reporter: mikeperry | Owner: mikeperry
Type: defect | Status: new
Priority: minor | Milestone:
Component: Tor-Tor server | Version: 0.2.1.22
Resolution: None | Keywords:
-----------------------------+----------------------------------------------
Old description:
> When the directory authorites vote that there are insufficient Exit nodes
> for them to be labled as both Exit and
> Guard, the authorities fail to add a bandwidth-weights line in consensus
> method 9. This has two problems:
>
> 1. We could actually create weights in may cases when this does happen.
>
> 2. This decision is made on the individual consensus *votes*, not the
> final consensus. There have been several
> situations so far when the authorites voted not to have Guard+Exit even
> though there was sufficient exit bandwidth
> in the consensus, and vice-versa.
>
> The problem is that this check probably is a good idea to perform,
> because of the large numbers of existing clients that
> we want to avoid using Exits-as-Guards with until they update to the new
> weighting system.
>
> The correct solution should be to tweak the Guard flag such that it
> produces a fixed amount of Guard-flagged nodes in
> the actual consensus. We should allow the weights themselves to handle
> what to do with Guard+Exit nodes.
>
> However, because this is a mostly harmless result that just causes
> clients to revert to the old weights when it
> happens, and because a bit of work and thought is needed to decide how to
> properly allocate the Guard flag, we're
> not going to just do a quick fix and allow weights for the D=0 situation.
> We should fix this issue once we feel
> that enough clients have upgraded to 0.2.2.x for it to make sense to fix
> it properly.
>
> [Automatically added by flyspray2trac: Operating System: All]
New description:
When the directory authorites vote that there are insufficient Exit nodes
for them to be labled as both Exit and
Guard, the authorities fail to add a bandwidth-weights line in consensus
method 9. This has two problems:
1. We could actually create weights in may cases when this does happen.
2. This decision is made on the individual consensus *votes*, not the
final consensus. There have been several
situations so far when the authorites voted not to have Guard+Exit even
though there was sufficient exit bandwidth
in the consensus, and vice-versa.
The problem is that this check probably is a good idea to perform, because
of the large numbers of existing clients that
we want to avoid using Exits-as-Guards with until they update to the new
weighting system.
The correct solution should be to tweak the Guard flag such that it
produces a fixed amount of Guard-flagged nodes in
the actual consensus. We should allow the weights themselves to handle
what to do with Guard+Exit nodes.
However, because this is a mostly harmless result that just causes clients
to revert to the old weights when it
happens, and because a bit of work and thought is needed to decide how to
properly allocate the Guard flag, we're
not going to just do a quick fix and allow weights for the D=0 situation.
We should fix this issue once we feel
that enough clients have upgraded to 0.2.2.x for it to make sense to fix
it properly.
[Automatically added by flyspray2trac: Operating System: All]
--
Comment(by mikeperry):
See also bugs #1206, #1115, and #1117, which suggest that maybe we should
just fix the WFU+MTBF code to always give us enough guards, and let the
balancing algorithms decide what to do with Exit flagged guards.
The reason this decision should be done in the consensus is that the
consensus weights are based on the actual final consensus. The decision to
not vote for Exits as Guards is currently being done with incomplete
information in the individual authority votes.
--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/1294#comment:1>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online