Re: QoS Solution?

[Roger Dingledine <arma@xxxxxxx>]

On Tue, Mar 29, 2005 at 01:11:23PM +0200, Michal Kolinski wrote:
> > I think we should dynamically decide fairness on a per-circuit basis
> > based on how many cells it's already tried to push through recently.
> 
> It won't work because many filesharing applications are using small (100-200 
> KB) chunks. Prioritizing small files would only prioritize filesharing apps 
> such as Gnutella clients or eDonkey2000 in "Horde" mode.
> The only fair QoS is equal bandwidth for every client. It works great for 
> example in NAT gateways, but it can't be used in Tor because someone can 
> open 10 Tor instances and get 10 times more bandwidth.

Hi Michal,

Both good points. But I think you're taking the wrong approach. We
shouldn't conclude that possible approaches "won't work" or "can't
be used" just because there exists a situation where they won't be
as effective.

I don't know of any magic answers here. No solution will work perfectly
for all situations. So we need to consider how much benefit we get from
each approach compared to how much complexity we add to the Tor code. I
think it might be worth trying this one out (especially as Mike keeps
volunteering to write code for us :)

--Roger