Idea

Calculate "forward loss" and "backward loss" heuristics and indicate these via new columns Floss and Bloss.

Definiton

"Forward loss" for a given probe P in a given round R indicates that (i) probe P was lost in round R and (ii) all subsequent probes within round R were also lost.

"Backward loss" for a given probe P in a given round R indicates that (i) probe P was lost in round R and (ii) a prior probe within round R has forward loss.

Informally, this is intended to indicate when a probe loss is the cause of subsequent losses ("forward loss") and when a probe loss is the victim of a previous loss ("backward loss").

Example

In the following example, after initially discovering the target (10.0.0.105) at ttl=5 in round 1, packet loss occurs in all subsequent round at ttl=3. No probes are able to get beyond ttl=3 and so we observe packet loss at hops ttl=4 and ttl=5. We can determine that the loss at ttl=3 is "forward loss" in that it causes all subsequent probes (ttl=4, ttl=5) in the same round to be lost. Likewise, we can determine that hops ttl=4 and ttl=5 have "backward loss" as they have loss in rounds where a prior hop (ttl=3) has "forward loss".

╭Hops──────────────────────────────────────────────────────╮
│#    Host         Loss%    Snd     Recv    Floss   Bloss  │
│1    10.0.0.101   0.0%     96      96      0       0      │
│2    10.0.0.102   0.0%     96      96      0       0      │
│3    No response  100.0%   96      0       95      0      │
│4    No response  100.0%   96      0       0       95     │
│5    10.0.0.105   99.0%    96      1       0       95     │

Note that in the above the Loss% column uses the traditional calculation ((Snd - Recv) / Snd). Similar columns could be added for Floss% and Bloss%.

Caveats

This technique is only a heuristic, it cannot conclusively determine forward and backward loss.

Every probe sent in every round is an independent trial, and so there is no guarantee that every probe within a given round will follow the same path (have the same "flow"). This invalidates the concept of forward and backward loss. We can therefore only meaningfully measure this if we can be confident all probs follow a single path. This is typically achieved (but not guaranteed) by tracing with the ICMP, UDP/Dublin or UDP/Paris strategy.

Even if a single path is followed by all probes, a given host on the path may drop packets for only a subset of probes sent within a given round, either due to rate limiting or genuine intermittent packet loss. This could result in a false positive "forward loss" for a given hop if all subsequent hops in the round showing loss which is not genuine (i.e rate limited). For example, hop with ttl=3 could be incorrectly be deemed to have "forward loss" for a given round due to observed loss from hops ttl=4 and ttl=5 which is not genuine (i.e. rate limited).

Similarly, a false positive "backward loss" could be recorded for a hop with genuine loss, if a previous hop shows loss which is not genuine (i.e. rate limited). For example, even if hop with ttl=4 has genuine loss, it will still be shown to have "backward loss" due to ttl=3.