Custom Query (11102 matches)
Results (16 - 18 of 11102)
|#1640||implemented||Combine torperf results from different sources||Karsten||karsten|
We have three TorPerf data sources that we're plotting in 3 graphs. We should combine these data in one graph, either by plotting median and 1st to 3rd quartile for all data points or by plotting three lines in a single graph.
|#1643||wontfix||Set up descriptor database for other researchers||Karsten||karsten|
There might be researchers who really want to work with our data but cannot set up their own database for some reason. Once we have a sane database schema with materialized views, we should ponder making a database available somewhere. This database might live on yatei, or better in the cloud. Ask Peter or Andrew how to provide a database to others without creating a security nightmare.
|#1644||implemented||Calculate request share from descriptor archives rather than believing in what relays say||Karsten||karsten|
When estimating the number of recurring daily users, we divide the number of locally observed requests by the expected share of requests seen at that directory. The result is the estimated number of total requests in the network. Right now, we're using self-reported request shares for this calculation, but these numbers may be wrong. Only recently in May and June 2010, the directories report very low shares of 1/10 of the usual shares or even 0.0%, which must be wrong. Rather than believing the relays here, we could determine request shares that directories should see from the descriptor archives. The descriptor archives reveal how many directory mirrors there are and what bandwidths they advertise that are used for load balancing by clients. The request share of one of the directories is its bandwidth divided by the bandwidth of all directories.