[Tor-BSD] FreeBSD tor-devel config options

George Rosamond george at ceetonetechnology.com
Thu Jan 8 15:53:58 EST 2015

Pete Wright:
> Hey All,
> I was doing some testing with my FreeBSD tor systems while poking around
> for potential optimizations.  I have found that setting the following
> compilation flags for the tor-devel FreeBSD port have lowered CPU load
> on my systems, and from the looks of things this may have lead to an
> increase of utilization of my circuit.
> enable: MEMPOOLS
> enable: THREADS [default]
> enable: TRANSPARENT [default]
> My guess is that enabling the MEMEPOOLS and BUFFREELISTS options has
> helped on my end.  I attempted to enable BUFFREEEVENTS as well, but that
> generated lots of logging data.  I may attempt to re-enable that once I
> have more data with my current configuration.

This is good stuff Pete.

The only question I have is the actual comparison of 'before' and
'after', as I've raised before.

If Tor was some static network it would be easy to test, but it's not.
So while the system might locally seem to perform better, it's still in
isolation to the public network that is randomly creating circuits.

The 'increased utilization' question is hard.  There are the normal
waves and troughs of the globe, ie, it's slower when it's 9-5 over the
Pacific Ocean, since that's the time most traffic is happening in any
particular locale from what I've been told.

I'm guessing we'd need a closed network with a directory authority and
all to really quantify this.... but that still doesnt deal with the
randomness of circuit selection.

Personally, CPU utilization never seemed to be the main weakness in Tor
performance, although I'm not discounting your tweaks as ineffective.

I'll check out your hax myself on NYCBUG0 which runs FreeBSD 10-stable.


More information about the Tor-BSD mailing list