[tor] Problems with high throughput Tor node

  • From: Moritz Bartl <moritz@xxxxxxxxxxxxxx>
  • To: torservers@xxxxxxxxxxxxx, tor-relays@xxxxxxxxxxxxxx
  • Date: Mon, 28 Feb 2011 10:38:09 +0100


CCed to tor-relays, the more appropriate list.

We should be careful not to mix the two, probably related, issues. Tor
needs to be improved for high bandwidth relays and multithreading. That
has always been the case.
NEW is the problem that currently, Tor does not use much bandwidth of
any of our exit relays.

On 27.02.2011 22:03, raidz wrote:
> I have a feeling the issue lies in the code, I mean Tor isn't even
> multi-threaded, thats pathetic.

The Tor developer team is a small group of excellent cryptographers.
Most of the work done on Tor is research based, and most research
doesn't care so much about real world issues like multithreading.

> I have a relay that maxes out at
> 100mbits on a gige line, it won't go any higher. It would be great if we
> could work together with the developers, possibly giving them access to
> some of our relays and exits to see if they can find where the
> bottleneck lies. 

As far as I can see they don't have anyone with special knowledge about
that, or the time to spend it. Their work is dependent on grants, and
those grants are for specific goals. Optimizing is not one of them.

I don't know about any bandwidth specific limitations in Tor. What I do
know is that Tor does not use multithreading, that's why we used as many
Tor processes as CPU cores to distribute load. It would be great to have
someone work on multithreading.
Also see

> With people getting more used to faster speeds it will
> turn some people off from using the network. I know speed isn't the
> first priority, but it should be one of the higher priorities.

I agree. And I hope that we can find solutions for that as soon as
possible, as we will probably ramp up some very high bandwidth exit
relays soon, and it would be a pity to waste [some of/most of] their

> I am not sure who would have better luck with the devs, I can try and
> reach out to them, although I know you probably have a better
> relationship with them. I would be more than happy to help them
> troubleshoot any issues, I just hope they have the time and are willing
> because as it is now, Tor is not scalable.

All we can do is hope for some help on the list, and bug devs on #tor.
:) Or, best, find some coder to work on it. The Tor project can use a
lot more coders, and all devs are very cooperative and helpful - once
they know you are really interested.

Moritz Bartl

Other related posts:

  • » [tor] Problems with high throughput Tor node - Moritz Bartl