[tor-relays] Relay bandwidth

[Roger Dingledine]

On Sat, Feb 08, 2014 at 10:41:21PM +0100, Andreas Krey wrote:
> On Sat, 08 Feb 2014 11:56:23 +0000, Tora Tora Tora wrote:
> ...
> > On a similar subject, is there a way to limit Tor's "per connection"
> > speed, i.e., not total speed.
> 
> No.

Actually there is:

       PerConnBWRate N bytes|KBytes|MBytes|GBytes|KBits|MBits|GBits
           If set, do separate rate limiting for each connection from a
           non-relay. You should never need to change this value, since a
           network-wide value is published in the consensus and your relay
           will use that value. (Default: 0)

       PerConnBWBurst N bytes|KBytes|MBytes|GBytes|KBits|MBits|GBits
           If set, do separate rate limiting for each connection from a
           non-relay. You should never need to change this value, since a
           network-wide value is published in the consensus and your relay
           will use that value. (Default: 0)

> > Assuming that a single connection carries
> > only one "conversation" between two parties at a time, wouldn't limiting
> > a single connection speed to, say 50-100Kb/s,
> 
> Actually, what would that be good for? As long as a relay is so lightly
> loaded that the active connections each can have more than than, there
> is no point in throttling them, and as soon as there isn't, they're
> fair-share-throttled down below that anyway.

I claim there is a point to throttling even when the guard node has
spare capacity. I explained it in:
https://blog.torproject.org/blog/research-problem-adaptive-throttling-tor-clients-entry-guards

And then two research groups answered it (in part) here:
http://freehaven.net/anonbib/#acsac11-tortoise
http://freehaven.net/anonbib/#throttling-sec12

But then see:
http://freehaven.net/anonbib/#pets13-how-low
for how all of these proposed fairness / throttling algorithms can be
trickier for anonymity than we first thought.

I'd like to turn on this per-conn static (not adaptive) throttling
network-wide:
https://trac.torproject.org/projects/tor/ticket/9368
but that's blocking on merging
https://trac.torproject.org/projects/tor/ticket/9762
but nobody has reviewed aagbsn's patch in #9762.

Whew,
--Roger