On Mon, 2008-10-27 at 17:05 -0400, Scott Lawrence wrote: > > Can loopdetectionkey be reduced to, say, 32 bits, rather than the full > > 128 bits? > > I'd prefer not to mess with that right now, since it could cause false > loop detections if two loops generated the same hash.
Looking at BranchId::loopDetected, loopdetectkey's are only compared within the branch parameters of an individual message. So every loopdetectkey is compared to maybe 20 other loopdetectkey's, which significantly reduces the chances for an accidental collision. Fiddling with the math, if there are at most V Via's per message, and we want to have at most probability C of a collision in each message, we need at least log2(V/C) bits in loopdetectkey. If V = 20 and C = 2^-32, we need at least 36.3 bits in loopdetectkey. If we use 66 bits in loopdetectkey (about 1/2 of what we do now), then C = 2^-39 = 10^-18, which seems big enough for practical purposes, given that there are only a few billion phone calls a day on Earth. So I propose that we abbreviate loopdetectkey to 66 bits (11 characters in base64). Dale _______________________________________________ sipx-dev mailing list [email protected] List Archive: http://list.sipfoundry.org/archive/sipx-dev Unsubscribe: http://list.sipfoundry.org/mailman/listinfo/sipx-dev
