Re: how much to use big relays (was Re: bloxortsipt)

[Roger Dingledine <arma@xxxxxxx>]

On Mon, Jan 04, 2010 at 08:23:28PM +0100, Olaf Selke wrote:
> Roger Dingledine schrieb:
> > You might instead ask about blutmagie, which at an advertised 41200 KB
> > is 3% of the Tor network.
> 
> Nevertheless I catch Roger's point. Is it a good idea exiting so
> much/many (never really understood the difference ;-)) tor traffic thru
> one exit gw? If it is not, I'm certainly going to reduce exit capacity
> by torrc config.

I think it's best to offer as much bandwidth as you want to offer, and
then let the directory authorities decide how much weight you should get.

For example, the bwauthority scripts currently compute how much weight
the directory authority suggests that we give to each relay:
https://svn.torproject.org/svn/torflow/trunk/NetworkScanners/BwAuthority/README.BwAuthorities
http://gitweb.torproject.org/tor/tor.git/blob_plain/HEAD:/doc/spec/proposals/161-computing-bandwidth-adjustments.txt
and they cap the vote they produce for any relay to 5% of the total
weights they're voting on. So while blutmagie varies between 1% and 5%
of the total network, it won't (I think) go higher than that.

The big challenge here is to find the right balance between security (not
too much centralization) and performance (making use of the bandwidth
that relays offer). One theory is "make sure to provide really good
decentralization, and one day when we have more relays it'll actually be
usable", whereas the other theory is "use what you've got as best you
can, and one day when we have more relays it'll be even safer". The 5%
heuristic tries to pick the right balance between the two. I'm more
willing to tend toward the second one now that we have active bw
measurement rather than passive "believe what the relay tells you"
measurement.

> https://torstatus.blutmagie.de/index.php?SR=Bandwidth&SO=Desc

For what it's worth, the numbers you see here aren't the numbers that
client use when weighting their path selection. You can see the weights
they use in your ~/.tor/cached-consensus file. It would be great if
somebody wants to patch the torstatus so it can reflect that.

More generally, here's a torstatus wishlist I wrote up a while ago:

  - Torstatus's relay listings should look in extrainfo descriptors and
    figure out the average actual bandwidth used by the relay. (Add up
    all the read and write histories, and divide by the number of seconds
    in the intervals.) That's the main bandwidth number it should show
    when ranking and sorting the relays. Maybe for completeness, we could
    add another column which just shows a number for the bandwidth the              relay is advertising.
  - Be sure to grab the advertised bandwidth from the consensus (the "w"
    line), not from the descriptor. Clients use the one in the consensus
    as of Tor 0.2.1.18.
  - I think torstatus still uses its own "tor check" variant. We should
    make it use the tordnsel instead (e.g. via fetching the list of
    current IP addresses from the bulk exitlist at
    https://check.torproject.org/cgi-bin/TorBulkExitList.py
    and comparing locally).

Volunteers happily encouraged. :)

--Roger

***********************************************************************
To unsubscribe, send an e-mail to majordomo@xxxxxxxxxxxxxx with
unsubscribe or-talk    in the body. http://archives.seul.org/or/talk/