wiki:org/meetings/2017Amsterdam/Notes/Metricsin5Years

Session notes from Fri, 24 Mar, 13:30: What could metrics and measurements look like in 5 years?

GATHER DATA

  • Deploy multiple ways to measure each thing (e.g. number of users), and analyze why they differ.
  • Graph analysis of circuit build failures and find network partitions.
  • Evaluate other GeoIP sources than MaxMind.
  • Add GeoIP data update mechanism, possibly after analyzing how often relays receive new GeoIP data right now.
  • Add client measurements: number of clients, number of mobile clients, mobility of clients, types of applications run by clients; without necessarily measuring on clients directly.
  • Separate the metrics module/process for metrics collection from the Tor daemon.
  • Automatically test if bundled bridges still work and tell the Tor Browser people.
  • Include Internet path measurements to and from Tor relays: performance, ASes traversed, countries on the path, etc.
  • Use the Tor process to record circuit build failures and detect partition attacks.
  • Look for rate limiting after a higher bandwidth. Might indicate rate limit confirmation.
  • Detect possibly benign anomalies: client using many guards simultaneously, uploading many descriptors to an HSDir, etc.
  • Measure abuse committed with Tor: port scans, SSH bruteforce attempts, exploitation of server vulnerabilities, copyright violations, etc.
  • Detect and report many types of attacks (bandwidth DoS, circuit building attacks, TCP packet spoofing, etc.)
  • Secure statistics aggregation with privacy guarantees (e.g. PrivCount).
  • Detect flood of HS descriptors per HSDir, look for disagreement between HSDirs.
  • Collect more types of benchmarks in OnionPerf (realistic web page models, etc.)
  • Plan for alternative Tor implementations and how they might not report statistics.

USE DATA

  • Users can create and share visualizations of Tor network data without writing a single line of code.
  • Create an automated system to detect anomalies in the Tor network (censorship, Sybil attacks) without all the false positives and in almost real-time, and automatically identify real-world events that they correlate with.
  • Include OONI data in queries together with metrics data.
  • Use OONI data on censorship and compare with detection data.
  • Use database for Onionoo that allows looking back in time.
  • Allows users to annotate events and provide annotations on website as "community-provided content".
  • Document what Tor network data exists and how it's being processes and analyzed: reproducible metrics.
  • Visualize Tor's security with respect to different adversaries.
  • Look for spikes in circuits created by guards.
  • Make all statistics robust against single false reports by relays or bridges.
Last modified 10 months ago Last modified on Mar 27, 2017, 6:44:50 PM