[tor-relays] Configuring SQM on nodes

[Mitar]

Hi!

Has anyone tested how does configuring SQM on Tor nodes influence the
quality of circuits over the node? Throughput, latency? So given that
circuits have 3 hops in serial I would assume that any
buffering/latency issues on nodes combine in a bad way. So using some
smart queue management (SQM) might improve things?

https://www.bufferbloat.net/projects/bloat/wiki/More_about_Bufferbloat/#why-does-sqm-work-so-well

So besides bandwidth measurements of nodes, should we also measure
latency contributions by nodes? I think this is the whole idea behind
the Bufferbloat movement, that it does not matter for user experience
the final throughput if you have bad latency. That what most users
perceive as "fast" is in fact good latency experience.

Tor have notions of sustained and burst bandwidth, but this to me
looks like just simplified QoS configurations. Shouldn't a better
configuration be "do not increase latency by more than X"? Then if you
hit the bandwidth limit, you are obviously increasing the latency more
because some packets are not getting through. But it also captures the
behavior better than "bursting is allowed up to Y".


Mitar

-- 
http://mitar.tnode.com/
https://twitter.com/mitar_m