14:30:17 <nickm> #startmeeting
14:30:17 <MeetBot> Meeting started Wed Sep 10 14:30:17 2014 UTC.  The chair is nickm. Information about MeetBot at http://wiki.debian.org/MeetBot.
14:30:17 <MeetBot> Useful Commands: #action #agreed #help #info #idea #link #topic.
14:30:20 <nickm> tor meeting, continues
14:30:26 <Sebastian> I guess I wanted to add that I was looking at bwauths
14:30:28 <nickm> hi Sebastian, what's up?
14:30:29 <nickm> ok
14:30:42 <armadev> ten points to sebastian for looking at bwauths. they are a mess.
14:30:52 <Sebastian> but I don't understand what's going on, and we didn't figure it out together either, and I'm sad about it
14:31:05 <Sebastian> then I got distracted by removing the naming code, which I want to finish first
14:31:06 <armadev> on the plus side, there are several research groups who want to work on secure bandwidth estimation. maybe they will solve 'accurate bandwidth estimation' while they're at it.
14:31:16 <Sebastian> but if anyone ever wants to do anything with bwauths, I'd be happy to help out
14:31:19 <Sebastian> if I can
14:31:52 <Sebastian> Here's my anaylsis:
14:32:01 <armadev> seems like 'have a spec for what the bwauth is trying to do' would be a nice first step. then somebody could write a clean bwauth.
14:32:22 <armadev> (and/or they could find flaws in the design)
14:32:30 <Sebastian> We're starting out measuring the network very quickly, bumping failing nodes back
14:32:48 <Sebastian> at some point, we only have nodes that are failing. And we keep trying them, and the timeouts are long
14:32:58 <Sebastian> which means we don't get the whole network scanned in even a day
14:33:43 <Sebastian> while we're running a scan, we don't take into account new information from Tor that we learn
14:34:13 <armadev> could we keep the long timeouts but start another test part-way in, on the theory that if x seconds have passed, you're probably going to end up timing about anyway?
14:34:15 <Sebastian> so after a short ramp-up time of quick progress (around 9% of the network / hour), we go down to measuring more like .x% of the network per hour
14:35:02 <Sebastian> So, the reason we don't do that is because at the end we need to assign weights, which is a relative thing
14:35:17 <Sebastian> if nodes keep coming and going it's harder to do that
14:35:37 <Sebastian> but yes. I feel if we had a spec someone could have a chance at implementing it sanely
14:35:57 <armadev> https://gitweb.torproject.org/torflow.git/blob/HEAD:/NetworkScanners/BwAuthority/README.spec.txt
14:35:58 <Sebastian> but I get lost because sometimes I wonder if I'm just not reading the code right or don't understand the design
14:36:06 <Sebastian> aagbsn and mikeperry have helped lots already
14:36:24 <Sebastian> yes, that document is outdated
14:36:29 <Sebastian> afaict
14:36:37 <armadev> good stuff
14:36:44 <Sebastian> ok, that's my report
14:36:53 <Sebastian> thanks for the people who helped me and for the review on the naming stuff
14:36:57 <armadev> also, the bwauth plan was to do a feedback loop, which we abandoned because it never worked the way the theory said it should
14:37:09 <armadev> so maybe we can simplify a lot by abandoning more of the plan that isn't working
14:38:12 <armadev> did we miss anybody else?
14:39:41 <nickm> it would appear not
14:39:49 <nickm> #endmeeting