[Author Prev][Author Next][Thread Prev][Thread Next][Author Index][Thread Index]

Re: Tor network speed



On Thu, Dec 30, 2004 at 08:11:41PM +0100, Andrew Henry wrote:
> If the minimum Tor server requirement is 160 KBits/s for all volunteer
> servers then 35 KBits/s is still quite slow.  Even if there are many
> hops between the entry and exit points, I would not have expected there
> to be such a big difference between that and normal router paths that
> may require several hops to a destination IP under normal circumstances
> without Tor.  Plus, I am assuming that Tor is not yet saturated by
> mainstream users.

Actually, it's quite possible that at various times various Tor servers
are saturated -- especially the ones in the lower bandwidth ranges, since
they're presumably used for other stuff as well and they don't have much
capacity to go around.

Weasel's bandwidth graph (http://www.noreply.org/tor-running-routers/)
indicates that we're pushing over 100 mbits/s average sustained throughput
over the Tor network, with bursts in the 200-300 mbit/s range.

The Tor software itself has been observed to push 10mbit/s on a given
node, and probably could do more if pressed. So the increased latency
(and decreased throughput) you're seeing is due to the servers your
circuit happened to pick.

> The reason for posting this question was to try to figure out if I can
> do some re-configuration to speed things up like specifying the entry
> point or something, for a server nearer to my location.  The second
> reason was to get some solid facts about the speed capability of the
> network, and to try to deduce if there is something that could be optimised.

Alas, as we try to scale the network to make it more robust and more
secure against various adversaries, we're going to start accepting
more and more slow nodes. This is inevitable, especially if we want
to avoid being such a juicy target for the Bittorrent folks -- see
http://archives.seul.org/or/talk/Dec-2004/msg00052.html

Some of this slowness is not inevitable. For example, we might later on
allow people to optimize paths through the network so they can choose
for themselves what tradeoff of performance vs privacy they want. Also,
I suspect our process of porting to Win32 is not yet complete, and the
Windows users are still getting a skewed view of what Tor can do. (The
core Tor developers don't use Windows, so we are relying on bug reports
from you folks to know what isn't working correctly yet.)

And finally, I've not given up hope on revamping the Tor innards to give
better quality-of-service to short streams (e.g. ssh) rather than big
streams (e.g. ftp). But that will wait until we've unearthed ourselves
from the latest flood of mails and feature requests. :) And in any case
it probably won't help you get your big web page faster, since it will
look (albeit only briefly) like an ftp stream.

Hope that helps,
--Roger