Simulation comparison - Feb 22nd 2016

[juniorz]

Graphs generated from 6fa2298 using $ ./verify

For a description of scenario and metrics, see this thread: https://lists.torproject.org/pipermail/tor-dev/2016-February/010418.html

The graphs are generated by collecting the stats from 500 independent simulations.

Default scenario

Original

Proposal

Success rates are close to 100% in both cases.

Fascist Firewall

Original

Proposal

Success rates are close to 100% in both cases.

Flaky network

Original

Proposal

Evil filtering

Original

Proposal

Success rates are close to 100% in both cases.

Sniper network

Original

Proposal

Auto switching between network scenarios

Original

Proposal

[juniorz]

After f20f854, there seems to be improvements in the success rate with minor impact on the exposure rate:

Flaky network

Proposal

Everything else looks pretty much the same, except for the Fascist Firewall (it might worsen its worst case scenario in terms of exposure):

Fascist Firewall

Proposal

[juniorz]

All the previous simulations were run choosing PRIMARY_GUARDS from REMAINING_UTOPIC at random, without taking into consideration the weighting by bandwidth. After acf6413 we use the bandwidth weights.

It seems to have improved some scenarios:

Fascist Firewall

Proposal

Flaky network

Proposal

Auto switching between network scenarios

Proposal