[Author Prev][Author Next][Thread Prev][Thread Next][Author Index][Thread Index]
Re: [tor-bugs] #23816 [Core Tor/Tor]: Use exponetial backoff with jitter and/or tune its parameters (was: Use expontial backoff with jitter and/or tune its parameters)
#23816: Use exponetial backoff with jitter and/or tune its parameters
-----------------------------+------------------------------------
Reporter: nickm | Owner: nickm
Type: defect | Status: needs_review
Priority: Medium | Milestone: Tor: 0.3.2.x-final
Component: Core Tor/Tor | Version:
Severity: Normal | Resolution:
Keywords: review-group-24 | Actual Points:
Parent ID: | Points:
Reviewer: catalyst | Sponsor: SponsorV
-----------------------------+------------------------------------
Comment (by catalyst):
I'm going to start by summarizing my comparison of the algorithms.
The existing algorithm is a weird hybrid of decorrelated jitter and equal
jitter. Each delay is at least the size of the preceding one, plus a
random value capped at a fixed multiple of the preceding delay. This
means in the worst case, d,,i+1,, = d,,i,, + d,,i,, * 3 + 1 = d,,i,, * 4 +
1, or approximately, d,,i,, = 4^i^.
For full jitter, the worst case is d,,i,, = base * m^i^.
For decorrelated jitter, the worst case is also d,,i,, = base *m^i^ but
any small variation in an early delay step magnifies in later steps.
The examples in the blog show m=2 for full jitter and m=3 for decorrelated
jitter. I'm not sure why they chose that variation.
The existing algorithm has some of the drawbacks of equal jitter, such as
missing some possible time slots because it tries to always include a
fixed delay component that's longer than the previous, so either
decorrelated jitter or full jitter would be an improvement.
--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/23816#comment:7>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online
_______________________________________________
tor-bugs mailing list
tor-bugs@xxxxxxxxxxxxxxxxxxxx
https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-bugs