Skip to content
Snippets Groups Projects
TODO.022 3 KiB
Newer Older
  • Learn to ignore specific revisions
  • Nick's initial priorities for Tor 0.2.2:
    
    NOTE 1: I'm not looking at fiddly little stuff from TODO.021 yet.  We
            can do a step where we triage the nice-to-have issues.
    
    NOTE 2: It's easy to list stuff like this with no time estimates and
            no target dates.  I think we should pick a target date for
            0.2.2, figure out how long the stuff we want will take, and
            triage accordingly, or vice versa.
    
    
      - Begin design work for UDP transition; identify areas where we need to
        make changes or instrument stuff early.
    
    - Performance, mostly protocol-neutral.
      - Work with Libevent 2.0's bufferevent interface
        - Identify any performance stuff we need to push back into
          libevent to make it as fast as we want.
    
      - Revise how we do bandwidth limiting and round-robining between
        circuits on a connection.
    
      - Revise how we do bandwidth limiting and round-robining between
        connections.
    
      - Better flow-control to avoid filling buffers on routers.
    
      - Split AES across cores if possible.
      - Split SSL across cores (reach; may require Libevent 2.1).
    
      - Figure out good ways to instrument Tor internals so we can tell
        how well our bandwidth and flow-control stuff is actually working.
    
        - What ports eat the bandwidth?
        - How full do queues get?
        - How much latency do queues get?
    
      - Rate limit at clients:
        - Give clients an upper bound on how much they're willing to use
          the network if they're not relaying?
        - ... or group client circuits by IP at the server and rate-limit
          like that.
    
    
    
    - Other features
    
      - Proposals to implement:
        - 146: reflect long-term stability
        - 147: Stop using v2 directories to generate v3 votes.
    
          - Start pinging as soon as we learn about a relay, not on a
            22-minute cycle.  Prioritize new and volatile relays for
            testing.
    
    
      - Proposals to improve and implement
        - 158: microdescriptors
    
        - 160: list bandwidth in consensus
          - and actually set it reasonably
          - and actually use it.
    
    
      - Proposals to improve and implement if not broken
        - IPv6 support.  (Parts of 117, but figure out how to handle DNS
          requests.)
        - 140: Directory diffs
        - 149: learn info from netinfo cells.
        - 134: handle authority fragmentation (Needs more analysis)
    
    
      - Needs a proposal, or at least some design
        - Detect client-status better
        - Have authorities report relay and voting status better: make it
          easy to answer, "Why is my server not listed/not Guard/not
          Running/etc"
        - Have consensuses come in multiple "flavours".
        - Weaken the requirements for being a Guard, based on K's
          measurements.
        - Adaptive timeouts for giving up on circuits and streams.
        - Have weight for picking new nodes as guards be less stupid
        - Lagged weight updates in consensuses: don't just move abruptly.
        d Don't kill a circuit on the first failed extend.
    
    - Installers
      - Switch to MSI on win32
      - Use Thandy, perhaps?
    
    
    - Deprecations
      - Make .exit safe, or make it off-by-default.