[tor-bugs] #7009 [Tor]: Handle unstable relays better
Tor Bug Tracker & Wiki
blackhole at torproject.org
Mon Jul 1 05:49:36 UTC 2013
#7009: Handle unstable relays better
---------------------------------+------------------------------------------
Reporter: arma | Owner:
Type: project | Status: new
Priority: normal | Milestone: Tor: 0.2.5.x-final
Component: Tor | Version:
Keywords: tor-relay, SponsorJ | Parent:
Points: | Actualpoints:
---------------------------------+------------------------------------------
Comment(by nickm):
Comments:
* This doesn't really have much to do with dynamic IPs.
* There's a lot of space going to relays that clients mostly don't use.
If we realizing that if we discarded all nodes with Bandwidth=X for X
under 100 we'd dump 48% of the nodes in the md consensus, but only about
half of a percent of the total capacity according to Bandwidth=X.
* If we decide to do that, though, we should re-run the
measurements.
* bzip2 is a clear win.
* For diffs, the "diff -e" output is (as predicted) a clear winner,
followed by a kludgey roll-our-own format.
* The main reason why diff -e wins is that my roll-my-own test
kludge format doesn't handle headers and footers so nicely. In quick-and-
dirty tests: if it were considering only the body, it would be within 3%
of diff -e.
* There still isn't a good C library we can link for diff. Otherwise,
we could do worse than exec as needed and require a working diff on every
Tor directory, I suppose.
* If we don't need so much bandwidth weight granularity, we could win
some space by dropping it.
* Rounding bandwidths off helps a little, but I'm not sure what we lose
by doing so.
* We don't actually know what fraction of clients download a new
consensus with what period. Sure, if you're online constantly, you get a
new one every 2-4 hours... but is that typical?
--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/7009#comment:10>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online
More information about the tor-bugs
mailing list