[tor-relays] bwadv vs bwhist + implications

[torhoste at tor.host-ed.me]

> hello there!
> very interesting information you shared here, thank you.
> while i dont have any further information to rely on,
> i would still like to share a few simple thoughts.
>
You're welcome.

> lets assume for a second that there are no errors in the code and that Tor
> is not using a significantly less than optimal network topology.
>
> i would argue against adjusting the traffic limit as you suggested, for
> the
> following reasons:
> -pending more data, this ratio might arbitrarily change at any point in
> time,
> causing either under utilization of the network. which is what you have
> set
> out to prevent.
> or over utilization which will cause troubles to the relay admins,
> like throttling by providers.
>
I clearly see where you're coming from. While I *certainly* do not propose
a large-scale adjustment in relay configurations (that isn't my incentive
actually), I still seriously think about doing the adjustments - certainly
with some offset to the observed figures to address potential
over-utilization - for our very own relays. We're looking forward to
receive some sponsored VPS(s) that we would like to dedicate/contribute to
the Tor network and I'm somehow concerned, i.e. think it's unfortunate,
that - expectedly - 1/3 of our monthly traffic limit(s) wouldn't be used
after all.

That said, please have a brief look at
https://metrics.torproject.org/bandwidth.png?start=2010-09-19&dpi=72&end=2012-09-19
- i.e. a two-year bwadv vs bwhist sample of the whole Tor network. It
actually shows a rather significant trend of the "1/3 bwadv vs bwhist
discrepancy" I observed previously.

> -its important to have more capacity then needed, this allows better
> stability and is helpful with dealing with sudden increases in Tor use, a
> somewhat common event,
>
Same here. I wouldn't want to argue against your assumptions as they -
from a general/common-sense perspective - make perfect sense. Still,
looking at the two-year sample graph, I cannot really see such a network
behavior, i.e. sub-samples where the gap between bwadv vs bwhist would be
significantly lower at times.

> of course, this is not to take away anything from your observation and
> initiative.
> who knows, you might have found a serious problem with the network.
>
I don't think there's a serious problem with the network. I just would
like to ask the list for the actual reasoning of the, from my perspective
rather *large*, discrepancy. You named three (actually two) things:

1. under-/over-utilization: the trend seems to be pretty constant and with
some offset calculated in, I would look forward to avoid over-utilization
(just for our own initiative certainly) and thus potential throttling by
providers for our very own contribution to the network.

2. adjustments to peak traffic: While I totally understand the argument, I
just can't see that happening (at least) within the last two years looking
at the sample graph.

thanks a lot for your feedback! Please don't get me wrong counter-arguing.
I'm just considering opportunities for our own initiative and clearly see
the general validity of your argumentation. Also I wouldn't want to
exclude the possibility that I'm missing something important here and
would like to ask everyone concerned to tell me where my (potential)
misconception actually is ;-)

Cheers,
Thomas

> thanks
>