Regulating bandwidth

Michael_google gmail_Gersten keybounce at gmail.com
Fri May 25 19:47:56 UTC 2007


May I strongly suggest keeping your bandwidth as high as you can
afford, even if it means the limit runs out early?

Right now the two biggest limits on Tor's speed is transmission speed
and CPU overhead. I'm realizing that there's no CPU limit -- there's
no "Only let N incoming routes exist" config parameter. And lowering
the transmission speed will only reduce Tor's speed.

Can we get a "CPU limit" / "Only let N incoming connections" parameter?

On 5/25/07, Roger Dingledine <arma at mit.edu> wrote:
> On Fri, May 25, 2007 at 10:59:07AM -0700, Andrew Del Vecchio wrote:
> > What's the best way to ensure that Tor spreads bandwidth more or less
> > evenly over a AccountingMax period? Currently, my 1GB of allocation
> > gets eaten up over the first three hours or so of operation.
>
> Currently we recommend that you manually set your BandwidthRate so
> your accounting quota is expected to be exhausted halfway through your
> accounting period.
>
> See the last paragraph of
> http://wiki.noreply.org/noreply/TheOnionRouter/TorFAQ#Hibernation
> for details.
>
> > I could
> > set the rate limits lower, but is there currently a way for Tor to
> > make inteligent decisions based on the average load? In other words,
> > in an ideal world, Tor would determine the average load, and normalize
> > it over the 24 hour AccountingMax period, giving the average rate as
> > the upper bandwidth rate during the day. This average would move a bit
> > over time, but always aim for AccountingMax over the course of the
> > average day.
>
> I wouldn't object to a config option AccountingBandwidthAuto or the
> like that does what you describe. This would save people from needing to
> do that calculation. In fact, the naive implementation could be really
> easy -- it could just take the AccountingMax and accounting period and
> do the math. But it could also try to get smarter, as you say.
>
> So, feel free to submit a patch. :)
>
> --Roger
>
>



More information about the tor-talk mailing list