Gregory Maxwell wrote: Unless you are editing through the SSL version. I've started doing that even though I don't use proxies, I don't think it's a bad idea for admins (at least) to use it. I've noticed very little difference in speed, certainly not enough to be bothersome.

On 9/8/07, Gregory Maxwell <gmaxwell(a)gmail.com> wrote: So don't ever log in to the non-secure one. I wonder if running Wikipedia as a tor hidden service would improve the speed. It'd at least spread the traffic more evenly as there'd be no need to find an exit node with an appropriate exit policy. 145.97.39.155 isn't the only ip address for en.wikipedia.org? How many are there?

On 9/8/07, Armed Blowfish <diodontida.armata(a)googlemail.com> wrote: These are LVS VIPs. I suspect that we could put in some sort of rewrite rules on the LVS hosts to redirect TOR traffic to some dedicated tor exit nodes which only allow traffic to reach back to the local LVS. I.e. to the outside world the TOR exits would look they are on 145.97.39.155 (knams), 66.230.200.100 (tampa), and 203.212.189.253 (yaseo), and 66.230.200.219 (secure). They would really be on other addresses. Their exit policies would allow traffic to :80 and :443 on their apparent external addresses. This should be enough to cause TOR to send all Wikipedia traffic to these exits. We could apply whatever blocking policy we want for TOR to the 3-4 actual exit source IPs. This would have the following advantages: 1) Less tor blocking inconsistency. (We often have only half the active Tor exists blocked from, which means that regular tor users can't edit via tor but sneaky trolls can... some exist are soft blocked, some are hard blocked, many are not blocked at all) 2) Improved security for users who use tor. No more risk of sniffing by naughty exit node operators. 3) Improved performance for tor users since there will be low latency between the exit and our caches. Even though allowing editing from Tor is a matter which rational people can debate... allowing people to read via tor is something we should support as strongly as possible.

On 0, Gregory Maxwell <gmaxwell(a)gmail.com> scribbled: But if you're using secure.wikimedia, would you be logged in on en.wikipedia? As I've said, I've been editing through the former for some months now, and whenever I am sent to an en.wikipedia.org link, I'm not logged in. Yes, that's definitely true. Every little bit helps though, and if you are running a server, you actually get really good browsing speed - subjectively, I felt like my Tor browsing no more than halved (or less on a good day!) the speed when I was running a server. I still don't know whether this is because it cut out a hop (since the server was now on the local machine) or because my traffic was somehow being favored. -- gwern security Ortega Firewalls propaganda 5707 b underground imapct Security Spall

On 08/09/2007, Gregory Maxwell <gmaxwell(a)gmail.com> wrote: Hey, I run an exit node, and I don't sniff the traffic. Note that the exit node can only sniff the IP of the middle node and the contents of the traffic itself - NOT the routing information of the client. This is assuming, of course, that the exit node operator is not performing a Sybil attack on Tor, i.e. running more than one node in the client's circuit. Entry node and exit node, combined with a latency timing attack against the middle node, should provide the equivalent of regular packet sniffing, but is much harder to do, especially if the client is using guard nodes, which substantially reduce the chance of this happening. Are you talking about session (cookie) stealing or password stealing? If an admin were theoretically using Tor, which is quite possible because all admins have ipblock-exempt, said admin could, as others have suggested, log in via TLS Wikipaedia. Of course, TLS has vulnerabilities, but unless the attacker is particularly determined and resourceful it should be good enough. Still, if you are worried, you could tell Tor to limit itself to a given set of exit nodes - preferably trusted ones. However, any random set of a limited number of exit nodes will reduce your probability of being sniffed, since rather than having to be any random exit node you happen to use, the attacker would have to be lucky to be one of the few exit nodes that you use. This would reduce your anonymity (which becomes pseudonymity if you log in to Wikipaedia), but it would also reduce the chance of sending your password by a malicious attacker. Changing one's password on a frequent basis is also a good security practise.