== Sponsor 4: Directory Bandwidth Reduction ==
Participants: karsten, n8fr8, ahf.
-
Alex went over the current efforts by the network team: compression scheme negotiation (proposal #278 (moved)), consensus diff (proposal #140 (moved)), and the proposals to make consensus diff documents smaller (proposal #274 (moved), #275 (moved), #276 (moved)) and the evaluation of compression algorithms1.
-
We should look into how our changes to compression algorithms, compression algorithm negotiation, and incremental updates ("consdiff") affects metrics. In particular, the network team and the metrics team (most likely Alex/Nick and Karsten) should stay in touch about this. The metrics have a current assumption that says that 10 directory connections per day equals one client, which this work might affect.
-
Will HTTP range requests make sense for directory downloads in countries with limited and/or flaky internet connectivity? It would allow little-t tor to fetch X bytes of a document, loose connectivity, fetch the rest Y bytes once connectivity is back? Too much troubles with file corruption where we have to ditch the file we just fetched or?
-
Are there any client fingerprinting issues related to doing any of this work?
-
Look into document fetch failure reasons in the metrics dataset. This should be possible with stem and collector.
-
Can we make bootstrap of Tor faster for mobile devices - especially for platforms where we cannot have processes running in the background?
-
n8fr8 explained how Facebook was doing the integration between Orbot and the Facebook client on mobile Android and which requirements that would be needed to do a similar thing on iOS: Shared library support.