Changes between Initial Version and Version 1 of Ticket #16861, comment 19


Ignore:
Timestamp:
Aug 27, 2015, 11:55:12 PM (4 years ago)
Author:
mikeperry
Comment:

Legend:

Unmodified
Added
Removed
Modified
  • Ticket #16861, comment 19

    initial v1  
    11FYI: My questions above still stand, but the following additional issues make me think that 0.2.8.x is a better target for this:
    2 1. Predictive circuit building is causing otherwise unused connections to live far longer than I expected in many cases (1.25-1.5 hours). This means that the overhead is more like 500KB per connection in the worst case, and even with ReducedConnectionPadding, 330KB in the worst case. I think that much overhead for mobile is not acceptable without allowing users to opt-out.
     21. Predictive circuit building is causing otherwise unused connections to live far longer than I expected in many cases (1.25-1.5 hours). This means that the overhead is more like 500KB per connection in the worst case, and even with ReducedConnectionPadding, 330KB in the worst case. I think that much overhead for mobile is not acceptable without allowing users to opt-out (or opt to reduce their connection lifespan).
    332. This probably also means we want the ability for clients to tell relays not to pad, or pad less, in case we discover that mobile connections should still live a long time, but pad less overall.
    443. I think all of this does mean we want statistics on average ORconn lifespan (issue 7 in comment:18), as well as stats on avg per-orconn padding, as this will help us tune the defense.