Hi Roger,
Thanks for getting back to me so quickly. I have read your email and
admit to being a little naive. For one I did not realise Tor is pushing
/ pulling 600mb per second !
I note your raised points about branding, and that I could be implying
its an official version, which I should not be doing in any case.
The reason for the 2 stacks is to enable hibernation of one without
affecting client behaviour, as you identified. It will be great when we
can both hibernate and still maintain client connectivity.
Based on your comments and those of Robert Ransom, I think its best to
cease work on this version and not distribute it. As you mentioned 2
stacks will result in double the descriptor traffic - which is not a
good idea. I was aiming this at home users who may have a small amount
of bandwidth to spare, but as you again mentioned, it appears this may
not help all that much, given the volumes in the network now.
Thanks for the comments and input. I am relieved I sought some feedback
before taking things any further.
With kind regards,
Cav Edwards
Roger Dingledine wrote:
On Thu, Oct 28, 2010 at 06:44:31AM +0100, Cav wrote:
We are just completing the 2010-11 release of Tor - Black Belt Edition.-
running on Windows.
Ah ha -- I remember seeing the "black belt edition" thing on a torrent
somewhere, and wondered who was doing it. Now we know -- great.
This version, with some tweaking, actually includes 2 vidalia+tor
stacks, running side by side.
One as a server, providing 15MB of bandwidth per day.
The other as an unrestricted client.
The use of the server, and its auto-starting, on Windows user-login is
controlled through the Start menu and install-time options. This is to
ensure that no user is 'forced' to run an exit relay without their
knowledge.
I am hoping for a wide exposure of this software version.
Just 1000 users of this version would be capable of adding 15GB !!! of
Server bandwidth PER DAY to the Tor network, with each user providing
their small amount. Surely this will help the Tor network immensely.
Four thoughts at first:
A) Why separate Tors and Vidalias? Tor is designed to be able to operate
as both a client and a relay at the same time. This is important because
every operating Tor fetches directory information over the course of
the day, and with hundreds of thousands of users, it adds up. Two Tors
means twice as much directory fetching load.
I guess one answer is that if you're using Tor's hibernation feature,
then the client functionality shuts off as well when Tor hibernates. I
just opened a trac entry to remind us about that:
https://trac.torproject.org/projects/tor/ticket/2129
B) 15MB per day is really not very much. First, notice that the Tor
network as a whole averages 600MB/s of traffic.
http://metrics.torproject.org/network.html#bandwidth
That's 52 TB a day. To double the capacity of the network, at 15MB per
relay, you'd need to get 3.5 million people running it. But it's actually
worse than that -- each Tor relay generates a server descriptor that
clients need to get in order to use it. So figuring a very conservative
100000 Tor clients running right now, and a conservative 500 bytes that
each user needs to fetch each day, that's 50MB that's going to be spent
just distributing the server descriptor to all the clients. So you really
want to be offering a few gigabytes per day before the numbers start to
become workable.
C) Many users are behind a nat that doesn't allow inbound connections.
Vidalia supports upnp, but I'm not sure how many users have routers that
support it. I wonder what approaches we can use to increase the chances
that users actually make sure their Tor relay is reachable.
D) The name "Tor black belt edition" sure implies that you're providing
an official Tor bundle. I wonder if there's a good name that doesn't
confuse users about who is making the package? See also
https://www.torproject.org/docs/trademark-faq.html.en
--Roger
|