Answer keys¶
For teachers. Every number here was measured by
web/test/labs.mjs
on the site's WebAssembly core, at the seed each sheet names. A student who
pauses a second early or late sees the last digit move, not the answer.
Lab 1¶
Sheet. MANET world, seed 6, at t = 60 s.
| run | delivered | ant transmissions |
|---|---|---|
| defaults (proactive on) | 89.0 % | 5559 |
enableProactive off |
89.0 % | 5210 |
- Prediction 1: delivery is the same. Most students predict a drop. That surprise is the lesson.
- Prediction 2: proactive ants cost about 6 % more transmissions here (5559 vs 5210).
- Explain 2: on this field and seed, nothing measurable. The reactive repair (link-failure notices and repair ants) already keeps up with walking-to-cycling speeds on the teaching radio.
- Explain 3:
--ns3::anthocnet::RoutingProtocol::EnableProactive=false. - Compare 1: proactive off is the measured MANET optimum: 97.2 % PDR at NRL 31.73, against 94.9 % / 40.51 at the 10 s default (#248).
- Compare 2: the default must serve every regime. On satellite constellations, sampling is what tracks the scheduled link changes (#192). A regime-specific setting is a configuration choice, not a new default (network-regimes.md).
Lab 2¶
Sheet. Mesh world, seed 8, at t = 120 s.
| run | delivered | ants per delivered packet |
|---|---|---|
| defaults (quiet mode on) | 99.6 % | 5.09 |
enableQuietMode off |
99.6 % | 5.21 |
- Prediction 1: about 5. The ns-3 static mesh reads 4.11 with quiet mode, 4.51 without.
- Explain 1: about 2 % fewer ants per packet in 120 s, with delivery unchanged. The saving is small because hellos and the initial route discovery dominate a short run. Quiet mode only thins the proactive share, and only once a route has held for a few rounds.
- Explain 2: it rarely engages. Every route change resets the back-off.
- Explain 3: a fully silent node would only notice a better route, or a failure the hellos miss, when data is lost. The back-off is capped (8 rounds) so sampling never stops.
- Compare: on the ns-3 static mesh, quiet mode cut NRL from 4.51 to 4.11 (−8.6 %) with delivery unchanged, and the default moved on in v2.3.0. It did not close the gap to OLSR's 1.78: the hellos (one per node per second) and the capped probing remain. That residue is the price of a protocol that would notice a change.
Lab 3¶
Sheet. Campaign level 3.4, seed 1. Delivery of all calls per 10-s interval, 0–160 s:
| run | 0–60 s, per 10 s | 60–160 s, per 10 s | mean 10–60 s | mean after 60 s |
|---|---|---|---|---|
| no roadside units | 0 0 0 23 0 50 | 30 0 3 5 50 8 0 0 0 0 | 14.5 % | 9.5 % |
| five units (par) | 0 33 50 73 90 100 | 100 23 40 50 85 57 0 0 35 5 | 69.0 % | 39.5 % |
- 3.1 quiz: the buildings on the corner block the line of sight. The cars are still within range.
- Explain 2: the units never move, so a car's link to one lasts as long as the car is on one of that unit's two streets. That gives the ants stable anchors.
- Explain 3: after 60 s only six cars remain. Some calls' endpoints are
out of reach of every unit for long stretches, so no route exists at all.
The
0 0intervals are partitions, not routing failures. - Compare 1: reconvergence. Routes break at corners faster than new ones are found and confirmed. 36 % of all traffic is lost to it on the ns-3 cell (vanet.md).
- Compare 2: a corner break is not a distance event. The two cars can be approaching each other when a building cuts the link. A predictor that sees only position and velocity predicts a long-lived link. That is the hypothesis recorded on #574 after its screen.
Lab 4¶
Sheet. Campaign level 5.3, seed 1. Delivery per 10-s interval, 0–120 s:
| run | per 10 s |
|---|---|
| the level (3 links cut at 40 s) | 60 75 65 45 70 60 40 75 70 45 75 65 |
| the same run, nothing cut | 60 75 65 45 75 70 40 75 65 45 75 70 |
- 5.1 quiz: each satellite sinks below 25°, and the station hands over to the next one rising.
- Prediction 1 and Explain 1: periodic. After the first interval the pattern repeats every 60 s, half the 120 s orbit. Halfway round, every satellite has moved to where its counterpart on the far side of the plane was, so the geometry the stations see repeats.
- Explain 2: about 20 s: the cut costs 5–10 points in the 40–60 s intervals. The node at the cut learns of it from its failed sends, and its link-failure notices tell the nodes upstream. The ants then re-weight the trails through the parallel paths, which already carry pheromone. After 70 s delivery is within 5 points of the uncut run.
- Explain 3: usually several. Between two satellites a planes and b slots apart, there are C(a + b, a) shortest staircase paths. Repair is a choice among paths the ants already know, not a new discovery.
- 5.3 quiz: it pauses briefly, then the ants find another path through the grid.
- Compare 1: OLSR learns of the failure through periodic topology control messages, then recomputes. Its reconvergence is paced by those intervals, not by the link.
- Compare 2: the interface-down fast path. When a point-to-point ISL fails, its interface goes down, and the adapter reports the neighbour lost at that instant.
Lab 5¶
Sheet. Mesh world, seed 4, at t = 120 s. The first call's source, data packets per next hop:
betaData |
next hops used | share of the busiest |
|---|---|---|
| 1 | 3 | 94.5 % |
| 20 (default) | 1 | 100 % |
| 30 | 1 | 100 % |
- Predictions: at β = 1 the source spreads, but only a little. One trail is far stronger than the others, and τ¹ still favours it. At 20 and 30, it uses one next hop.
- Explain 2: more paths spread load and survive a single failure. The cost is packets arriving out of order and some taking longer paths, so delay and its tail grow.
- Explain 3: the thesis favours delay and in-order delivery. Multipath stays available, and the forwarding rule uses it when trails are close in strength.
- Compare:
heavy-load, on delay99: −192.50 ms [−210.65, −174.40] when β₃ = 20 was adopted (#179).