Changes between Initial Version and Version 1 of org/meetings/2018MexicoCity/Notes/TorPerformance


Ignore:
Timestamp:
Oct 7, 2018, 4:16:54 AM (5 months ago)
Author:
arthuredelstein
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • org/meetings/2018MexicoCity/Notes/TorPerformance

    v1 v1  
     1Will aim to cover anything that slows Tor down
     2
     3https://arthuredelstein.net/exits
     4
     5https://metrics.torproject.org/torperf.html
     6data of individual relays -- collector
     7
     8
     9https://collector.torproject.org/archive/torperf/
     103 file sizes: fresh download each time
     11 plan: download a single file
     12 
     13 https://metrics.torproject.org/onionperf-buildtimes.html
     14 
     15 
     16 Putting metrics into relay code:privcount
     17  Buffer
     18  DNS failures
     19  Could add things to extrainfo
     20 
     21  Tor over QUIC as a use case
     22    DTLS?
     23    Does metadata get leaked?
     24    QUIC has pluggable congestion control mechanisms
     25    https://www.benthamsgaze.org/2016/09/29/quux-a-quic-un-multiplexing-of-the-tor-relay-transport/
     26    (This architecture considers QUIC hop-to-hop, not end-to-end for Tor)
     27   
     28IPv6 work in Tor
     29
     30Enabling ECN on relay hosts may reduce head of line blocking on inter-relay connections
     31
     32Slow relays make the network slower. On slow relays you see large queues.
     33
     34Early privcount statistic
     35
     36
     37Changing topology: 100x factor: not every relay can talk to every other relay. Restricted topologies.
     38
     39Shadow can simulate packet loss and jitter, but only TCP connections.
     40
     41Experimenting with different protocol stacks is getting easier.
     42
     43Alternative link protocols would be comparitively easy to switch out.
     44
     45Sidechannel issues.
     46
     47Can point-to-point links improve performance? Maybe DTLS?
     48
     49Different route selection protocols. More direct route crosses fewer jurisdictions. If path selection depends on destination, then it helps adversary reduce number of destinations.
     50
     51With privcount we might be able to get statistics over every 4 hours.