Isn't that an instructive glop of information? Here's what it means,
using the top line:
The set of paths containing 0 vbns links out of 5 total links (which
is 0.0% VBNS infection for this grouping) contained 165 members.
Which? What? Huh? We're talking about the result of running shortest path
over the runtime graph. "5 links" means that there was a shortest path
value, chosen as that of the best performance, that consisted of five
links, between two nodes. (We're not doing the triangles any longer, at
least not exclusively).
How did we choose the paths? Although the sampling done "online" during
the initial measurement phase was killed due to a graph error, the
time-average measurements survived intact, and were fed into a repaired
graph. This graph was then left to sample at hyperspeed, overnight, just
like it samples during runtime. (But without making measurements).
Looking at this a bit more, we find that the sum of the number of samples
is:
165 + 115511 + ... + 76 = 151106 total samples.
And
92 + 3058 + 76 = 3226.
Thus, overall, 3226/151106 = 2.135% of our alternate paths traverse any
VBNS links. As one would expect, these are concentrated around 2 hop
alternates. As one would not expect, there are no 2-VBNS link paths.
Rechecking the code confirms its correctness, at least with the eyes of
now.
Measurements are ongoing that do not place any special emphasis on VBNS.
_________________________________________________________________________
"A government big enough to give you everything you want is big enough to
take everything you have."