=== Research related follow up questions- Network team meeting 2019 ===
-
Conflux
a. How would retransmission work in the case where one circuit is destroyed? What load would this place on the exit for buffering cells and is this load acceptable?
b. How would Conflux work with authenticated sendmes?
c. Could Tor use a dynamically-sized buffer for sendmes deliver/package window, instead of the fixed 100 cells as is currently? -
Walking onions
a. Can we create a circuit building model that doesn't rely on telescoping?
i. Can this be done in a way that also allows for forward secrecy and unlinkability?
ii. Can this be done in a way that hides the random choice from the guard/middle node, but doesn't allow the guard/middle node to influence the choice in a way that is unsafe? -
WTF-Pad
a. Verification that this in fact breaks the deterministic attack against onion circuits (cite paper)
b. We need new better models- the current models can be optimized
c. Understand how this will impact the network when it is under congestion
d. Better understanding of how this will work with low-bandwidth devices such as mobile
e. Adversarial machine learning for building new models? -
Privcount
a. What consistency checks are further need?
b. How to build more complex counter types than a simple counter, such as a HyperLogLog counter. -
Vanguards
a. Validate this fixes the attack in Florian's paper
- Bandwidth scanner
[Feel free to convert on "real" research questions]
The main one is:
- What should be the Tor network load balance goal? [0].
We are scaling the raw bandwidth values in the same way torflow does. The raw bandwidth distribution is almost linear, after scaling the distribution becomes almost exponential. We don't know which should be the ideal distribution
Other open questions (many of them collected in Rome meeting):
- What if the scanners would report the raw bandwidth and directory authorities would scale the bandwidth?
- What should be the relation between bandwidth and consensus weight?
- How to increase network diversity?
- how to distribute better network load around the world?
- what balance do we want between optimizing performance and maintaining network security?
- How the location of the scanners affect the measurements?
- How the location of the Web servers affect the measurements?
More research questions where the bandwidth scanner design would be different:
- "proof of storage" protocol, which delegates the bandwidth measurements to little relays, which run in aggregate to measure bigger relays
- Scanner traffic not fingerprintable (peerflow)
- Can we use conflux or something like this? it could be a DDoS against the new relays. This could be done selectively against those relays with the unmeasured flag.