Forward Loss Threshold

[fujiapple852]

Currently, forward loss for lost probe P is detected by examining if all subsequent probes are also lost, provided there is at least 1 subsequent problem.

This can lead to false positive for the 2nd last probe if the final probe appears "lost" but is not (i.e. is rate limited). The probability of this happening is high as it only takes 1 probe (the final probe, to the target) to be "lost" for forward loss to be declared. For the 3rd last probe it needs both the 2nd last and the last probe to be lost before it is considered forward loss.

For example, if hop 3 has a 10% chance of showing "fake" loss (i.e. is rate limited) then the probability of hop 2 incorrectly show forward loss 10% of the time:

TTL	State	Fake Loss %
1	Complete	0%
2	Lost	0%
3	Lost	10%

Here, if hops 3 and 4 both have a 10% chance of showing "fake" loss then the probability of hop 2 incorrectly showing forward loss is reduced to 1%:

TTL	State	Fake Loss %
1	Complete	0%
2	Lost	0%
3	Lost	10%
4	Lost	10%

Therefore, it may be sensible to set a minimum threshold for the number of subsequent hops that must show loss in a given round before it is deemed forward loss. Currently this minimum is implicitly fixed at 1, whereas it could be made configurable and given a larger default value of at least 2.