RIPE69 Brainstorm

Some key points from the Thursday morning brainstorm we had about OpenIPMap

• Create better incentives to crowdsource

  • Crowdsourcer of the week (and call him/her Hercules (because of the cleanup)?) http://en.wikipedia.org/wiki/Labours_of_Hercules#Fifth_Labour:_Augean_stables
  • give RIPE Atlas credits as incentive
  • Mechanical Turk-approach: give "users" 10 new traceroutes to work on (every day); let the user chose the criteria for these: UDMS targeting my country, my city, UDMs going through (my/specific) ASN...
  • a way of selecting all traces going through specific AS / region

• "MOST WANTED" scoreboards: Stats on how well we cover certain cities / networks in openipmap, so we can see where to improve

• Side discussion on RIPE atlas coverage

  • Viz of ripe atlas coverage
  • For top 100 cities (by population): how well are we doing getting probes there / how far away
  • GDP per capita vs probe density map
  • Targetting specific locations not already covered helps OpenIPMap, specifically latency-based measurements

• hostname schemas

  • We briefly touched this topic, discussed using restriction based on ASN/hostname combinations.

• RTT constraints

  • MPLS is a serious issue. Having a direct path (in traceroute) from LA to Djibouti, that path is probably going to go through various hops in MPLS that are not seen (Amsterdam? London?) for common paths/RTTs users could crowdsource the paths this goes through, if they have high confidence about these.

• Anycast

  • At least there needs to be a way to flag anycast prefixes, and this info needs to trickle down through all parts of the system (API/viz)

• Mapping ip-pairs to cable-paths

  • Because a lot of the cable-paths are very similar instead of putting straight lines (or great arc) on a map, these could be crowdsourced to well known cables (http://www.cablemap.info/ , or Telegeography). Sometimes the carrier (identified by hostname or ASN) is associated to a specific cable, in which case this is easy. In other cases there are common paths that multiple cables take, so we could use an abstraction of common paths for that. Also a lot of terrestrial fiber paths are open data, so these could be better mapped too.

• High PR value: “beauty contest” - selection of prettiest traces; “before & after” pairs