Meeting Notes - SponsorR on #tor-dev Date: 2014-11-18, 16h00 UTC
Attendees: David Goulet (dgoulet), George Kadianakis (asn) and Karsten Loesing (karsten)
It started by George mentionning that this meeting would be SponsorR general meeting and also a discussion on the statistics we need to gather for the sponsor on the real Tor network.
Relevant document for the meeting: https://etherpad.wikimedia.org/p/hs_stats_78281091
Status report followed:
Focus on statistics analysis to answer two questions:
"how much HS bw"/"how many HSes"
A proposal is being written (35% completion) that can be found here https://pastee.org/6e9dr. Should be finished by next tuesday and next week the code should start.
So on the proposal George is still unsure on two things:
a) About the "How much HS bw" part. Should relays post number of cells or number of circuits or both"? And should they only send number of cells/circuits on RP circuits, or on RP+IP circuits?
b) How much should we obfuscate the "How many HSes" statistics? I'm going for 10% noise. Which means that if an HSDir sees 10 HSes, it will report either 9 or 10 or 11.
The discussion on that was move for later on Karsten suggestion.
Have a chutney network with hidden service support that is ready to use with Torperf for measurements (code instrumentation to come).
In Boston for the week visiting nickm for mentoring, collaboration, and help on the Tor code base.
Proposes to collect mostly timing measurements and stats that are in the pad for the privnet. (info to be added to #13792)
Next step is to instrument Tor code base to gather the information we need.
Spent time on the proposal draft with George who suggested that we should send that asap to tor-dev mailing list.
< karsten> so, I think we have a fine collection of stats there. some of which we'll write a proposal for, but not for all.
With those stats in the pad (see document at the beginning), most of them will go in the proposal and some will be dropped.
Proposed that we turn this proposal (pad stats) in a tech report draft instead which was accepted by all attendees. We keep writing the proposal that George worked on but for the rest (other stats) we go towards a tech report.
A discussion followed on the HS stats that covers the two questions George is trying to answer:
< asn> a) About the "How much HS bw" part. Should relays post number of cells or number of circuits or both"? And should they only send number of cells/circuits on RP circuits, or on RP+IP circuits?
Karsten argued that cells would be more accurate than circuits.
<karsten> Though circuits would also work as first approximation. What we should do is combine these stats with cell stats.
George is also leaning towards cell count for a better idea on the total bw. Karsten informs that cell stats is already being gathered by some volunteer relays.
V(CellStatistics, BOOL, "0"),
Karsten suggested that we design HS bw stats in a way that we can later combine them with cell stats. George seems uncomfortable at this point and prefer to be as safe as we can instead of making sure our sponsor is happy because this could leak too much info and might have disastrous consequences.
< asn> that said, "fraction of total traffic" is also interesting for load balancing reasons < asn> for example, if we want HSes to have different path selection algorithm than the other clients.
Finally, Karsten proposes this which was accepted:
< karsten> we only do cell stats for HS purposes now. we turn on cell stats on the volunteer relays. we compare how accurate our stats are for computing fractions without cell stats. < karsten> we hope we don't need cell stats turned on, or cell stats for normal purposes in the HS stats line.
George asked if Karsten could take over the proposal next week which was answered by yes and we should meet again next week of course.
George will gather all information on the SponsorR thread on tor-dev to put it in the wiki: