[tor-bugs] #20459 [Core Tor/Tor]: ewma_cmp_cmux never considers policies different
Tor Bug Tracker & Wiki
blackhole at torproject.org
Thu Nov 3 18:58:01 UTC 2016
#20459: ewma_cmp_cmux never considers policies different
-------------------------------------------+-------------------------------
Reporter: pastly | Owner:
Type: defect | Status: closed
Priority: High | Milestone: Tor:
| 0.3.0.x-final
Component: Core Tor/Tor | Version: Tor:
| 0.2.6.2-alpha
Severity: Normal | Resolution: fixed
Keywords: 029-backport, review-group-11 | Actual Points:
Parent ID: | Points:
Reviewer: | Sponsor:
-------------------------------------------+-------------------------------
Comment (by pastly):
Ready for some graphs? This data was collected from two shadow
simulations. Each has 500 relays and 7500 clients. The one line summary of
the results (to me) is: the ewma fix does no harm by the metrics I
gathered.
qtime.shadow.results.pdf shows the amount of time a cell spends in the
kernel outbound buffer after leaving Tor and becoming bytes. As you can
see, no change.
shadow.results.pdf shows a lot.
- Time to download `x` bytes are for each different type of client.
Clients with a smaller `x` behave a lot like web browsers. Clients with
larger `x` are near continuous bulk downloads. Looks like no change.
- A bunch of probably self-explanatory graphs all showing no change.
- 60 second moving average read (pg 22) and write (pg 25). We give the
network 30 simulation minutes for every relay/client to boot up and reach
steady state, then measure for 10 simulation minutes. 'After' seems more
stable than 'before', which would explain the better read and write
performance for 'after'.
So, IMO, the fix causes no harm and is beneficial. I can answer any
questions about what graphs mean, what the simulations consist of, etc.
I will attach the files right after hitting submit on this. Because I
can't do it all at once ... ?
--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/20459#comment:14>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online
More information about the tor-bugs
mailing list