Feature/mobility handover sigma

.github/workflows/deploy-docs.yml

@@ -9,6 +9,7 @@ permissions:
   contents: read
   pages: write
   id-token: write
+  administration: write
 
 concurrency:
   group: deploy-docs
@@ -31,6 +32,8 @@ jobs:
         run: zensical build
 
       - uses: actions/configure-pages@v5
+        with:
+          enablement: true
 
       - uses: actions/upload-pages-artifact@v3
         with:

docs/mobility-formal-model.md

@@ -121,7 +121,7 @@ PSA-region crossing.
 
 $$\sigma_{5G}^{\mathrm{PSA\text{-}reloc}} \approx 1500\ \text{bytes}\quad\textit{(approx — to ground in TS 23-502 §4.3.5; SSC 2/3 only)}$$
 
-### 3.5 SSC-1 anchor path-stretch (the real intra-PLMN cost of pinning)
+### 3.4 SSC-1 anchor path-stretch (the real intra-PLMN cost of pinning)
 With the anchor pinned, 5G traffic hairpins from the current serving edge UPF to the
 **original** PSA; the optimal egress (what 6G-RUPA achieves by renumbering into the
 local domain) is the **nearest** PSA. The excess
@@ -132,7 +132,7 @@ is exactly *why* SSC-1 pinning is acceptable intra-PLMN. It grows large only whe
 anchor is far from the user: **roaming** (hairpin to the home country) and edge-compute
 (motivating SSC 2/3). Measured in-sim: `anchor_dist_5g_sum` / `anchor_dist_opt_sum`.
 
-### 3.4 6G-RUPA — renumbering (all levels, flat)
+### 3.5 6G-RUPA — renumbering (all levels, flat)
 **Mechanism (Grasa et al. 2017 §III; RINA RM lines 1814–1828):** the moving IPCP
 (UE) obtains a new synonym (address) reflecting its new location and then:
 1. **advertises the new address via routing** to its direct neighbours
@@ -166,7 +166,7 @@ $$\sigma_{\mathrm{RUPA}}^{\mathrm{renumber}} \approx 200\ \text{bytes (flat acro
 > domain's aggregate prefix is fixed by topology and already present; the UE merely
 > adopts an address under it. No core prefix is added or removed at any level.
 
-### 3.5 Roaming
+### 3.6 Roaming
 - **5G Home-Routed** (inter-PLMN): visited UPF anchors back to the home PLMN over
   N9; per-roaming-session inter-PLMN coordination.
   $\sigma_{\mathrm{roam}}^{5G,HR} \approx 1180$ B [TS 23-501 §4.2.8.2.3; TS 29-244 §8].
@@ -292,6 +292,62 @@ HR roaming (separate scenarios). Granularity equality holds regardless.
 
 ---
 
+## 6b. Service continuity & in-flight packet loss (the "what about packets in flight?" result)
+
+**Claim: 6G-RUPA renumbering loses no in-flight packets at the core; 5G's
+tunnel teardown/re-establish opens a loss window on anchor-affecting handovers.**
+
+Decompose a handover into a **radio leg** (RRC reconfiguration + random access to the
+target cell) and a **core leg** (data-path update):
+
+- **Radio leg is architecture-agnostic.** 5G and 6G-RUPA use the *same* NG-RAN; the
+  radio interruption time, radio-link-failure and ping-pong behaviour are identical
+  and **out of scope** for an addressing-layer comparison. (This is exactly where the
+  WCNC experiments saw their only loss — an IRATI↔WiFi WPA-supplicant artifact, not
+  the renumbering — `rina-mobility-wcnc.md` §IV.)
+- **Core leg is where the architectures differ.**
+  - **5G:** an anchor-affecting handover (L2 N2/UL-CL reloc, and SSC 2/3 re-anchor)
+    **tears down and re-establishes UPF tunnel state** (PFCP Release/Establish). During
+    that window in-flight packets on the old path are lost or reordered unless masked
+    by data-forwarding/end-markers; an L3/SSC-2 break-before-make additionally changes
+    the UE IP. Xn (L1) is the mild case (end-marker forwarding).
+  - **6G-RUPA:** the flow is **never torn down** — the address is a *synonym*, EFCP is
+    keyed on port-ids, and the renumber is **make-before-break** (new synonym advertised
+    + flow-update sent to correspondents *while the old address still forwards*, retired
+    only after a policy timeout; RM l.1814–1828). So **ΔS_ctx and the flow survive with
+    zero core-side loss** — *"there are no tunnels to set up and tear down; packet loss
+    can only occur if there was no physical communication"* (`rina-mobility-wcnc.md` §III;
+    0-loss renumbering, Grasa et al. EUCNC 2017).
+
+$$\text{in-flight core loss} = \begin{cases} >0 \text{ window} & \text{5G anchor-affecting (tunnel re-establish)} \\ 0 & \text{6G-RUPA (make-before-break renumber)} \end{cases}$$
+
+We argue this **architecturally** (not via per-packet simulation): the result is a
+property of *whether the data path is torn down*, which the σ/state model already
+captures (5G re-establishes per-session state; RUPA does not).
+
+## 6c. Metrics mapping — what the literature measures vs what we isolate
+
+Standard 5G handover KPIs (surveys): handover interruption time (HOIT), handover
+delay/execution time, **packet loss**, throughput, **signaling overhead**, handover
+rate (HOR), success/failure rate (HOSR/HOFR), radio-link-failure (RLF), ping-pong.
+
+| KPI | Layer | In this paper? |
+|---|---|---|
+| Signaling overhead (σ) | core/CP | **Yes — central** (Xn/N2 vs flat renumber) |
+| Forwarding/core-state size & churn | core/UP | **Yes — headline** (our contribution; not a classic mobility KPI) |
+| In-flight loss / service continuity | core/UP | **Yes — architectural** (§6b: 5G window vs RUPA 0) |
+| Handover rate (HOR) | radio geom. | **Yes** (validated vs Voronoi geometry) |
+| Anchor path-stretch | core/UP | **Yes** (§3.4; our addition) |
+| HOIT, HOD, RLF, ping-pong, throughput | **radio/RAN** | **No — architecture-agnostic** (same NG-RAN), out of scope |
+
+The framing for reviewers: classic mobility KPIs are dominated by the **radio** leg,
+which is identical for 5G and 6G-RUPA; we deliberately **isolate the core/addressing
+leg**, where the architectures actually differ, and contribute two metrics the radio-
+centric literature does not track — **core forwarding-state churn** and **anchor
+path-stretch**.
+
+---
+
 ## 7. Summary table
 
 | Handover | 5G σ (B) | 5G ΔS_core | 5G ΔS_ctx (billing) | RUPA σ (B) | RUPA ΔS_core | RUPA ΔS_ctx |

gen_trajectories.jl

@@ -5,7 +5,8 @@
 # under the chosen mobility model, and records its path + handover events
 # (classified L1/L2/L3). Writes viz/data/{trajectories,gnbs,meta}.json.
 #
-#   julia --project gen_trajectories.jl [n_agents] [duration_s] [dt_s] [speed_kmh]
+#   julia --project gen_trajectories.jl [country] [n_agents] [duration_s] [dt_s] [speed_kmh]
+#     country = spain | usa | usa_asr   (default spain; agent count defaults to national)
 #
 # No full DES needed: we just iterate step_position + find_serving_gnb per agent,
 # exactly the per-tick handover check Core.jl does, so levels match the national run.
@@ -14,18 +15,34 @@ using DesJulia6gRupa, DesJulia6gRupa.Types
 import DesJulia6gRupa.Simulation as DSim
 import DesJulia6gRupa: select_agent_location
 
-const NAG      = length(ARGS) >= 1 ? parse(Int,   ARGS[1]) : 1500
-const DURATION = length(ARGS) >= 2 ? parse(Float64,ARGS[2]) : 600.0
-const DT       = length(ARGS) >= 3 ? parse(Float64,ARGS[3]) : 4.0
-const SPEED    = length(ARGS) >= 4 ? parse(Float64,ARGS[4]) : 50.0
+# (data subdir, gNB csv files rel to data/<sub>, operator net id, #edge UPFs, population)
+# Mirrors run_national.jl PROFILES so trajectory levels match the national run.
+const PROFILES = Dict(
+    "spain"   => ("spain", ["opencellid/214.csv"],                      7,  52,  49_442_844),
+    "usa"     => ("usa",   ["opencellid/310.csv","opencellid/311.csv"], 480, 817, 335_000_000),
+    "usa_asr" => ("usa",   ["asr/310.csv"],                             999, 817, 335_000_000),
+)
+const SCALE    = 1000
+const ADOPTION = 0.82
+const NUM_PSA  = 5
+
+const COUNTRY  = lowercase(get(ARGS, 1, "spain"))
+haskey(PROFILES, COUNTRY) || error("unknown country $COUNTRY (spain|usa|usa_asr)")
+const SUB, FILES, OPID, NEDGE, POP = PROFILES[COUNTRY]
+const NAG      = length(ARGS) >= 2 ? parse(Int,   ARGS[2]) : ceil(Int, POP*ADOPTION/SCALE)
+const DURATION = length(ARGS) >= 3 ? parse(Float64,ARGS[3]) : 600.0
+const DT       = length(ARGS) >= 4 ? parse(Float64,ARGS[4]) : 4.0
+const SPEED    = length(ARGS) >= 5 ? parse(Float64,ARGS[5]) : 50.0
 const NSTEPS   = floor(Int, DURATION / DT)
 
 r5(x) = round(x, digits=5)      # coord rounding to shrink the file
 
-println("Building Spain topology (52 edge / 5 PSA, two_tier)...")
-paths = filter(isfile, [joinpath(@__DIR__,"data","spain","opencellid","214.csv")])
-topo = DSim.load_and_deploy_network(paths, 7, 52, joinpath(@__DIR__,"data","spain"),
-                                    SimConfig(1,2,1000,1,1,1,:two_tier,5,1))
+println("Building $(uppercase(COUNTRY)) topology ($NEDGE edge / $NUM_PSA PSA, two_tier)...")
+base = joinpath(@__DIR__, "data", SUB)
+paths = filter(isfile, [joinpath(base, f) for f in FILES])
+isempty(paths) && error("no gNB data under $base for $FILES")
+topo = DSim.load_and_deploy_network(paths, OPID, NEDGE, base,
+                                    SimConfig(1,2,SCALE,1,1,1,:two_tier,NUM_PSA,1))
 println("gNBs=$(length(topo.gnb_locations)) edgeUPF=$(length(topo.upf_locations)) PSA=$(length(topo.centralized_upf_locations))")
 
 model = RandomWaypoint(SPEED, 0.0, SPEED*DT/3600*2)   # max jump ~2 steps of travel
@@ -74,27 +91,27 @@ end
 return nho
 end
 
-io = open(joinpath(outdir, "trajectories.json"), "w")
+io = open(joinpath(outdir, "trajectories-$COUNTRY.json"), "w")
 print(io, "[")
 nho = gen_trips(io, topo, model)
 print(io, "]")
 close(io)
-println("Wrote trajectories.json  ($NAG agents, $nho handovers)")
+println("Wrote trajectories-$COUNTRY.json  ($NAG agents, $nho handovers)")
 
 # gNB sites (rounded). 46k points render fine in deck.gl.
-open(joinpath(outdir, "gnbs.json"), "w") do f
+open(joinpath(outdir, "gnbs-$COUNTRY.json"), "w") do f
     print(f, "[")
     for (i, g) in enumerate(topo.gnb_locations)
         i > 1 && print(f, ",")
         print(f, "[", round(g.lon,digits=4), ",", round(g.lat,digits=4), "]")
     end
     print(f, "]")
 end
-println("Wrote gnbs.json  ($(length(topo.gnb_locations)) sites)")
+println("Wrote gnbs-$COUNTRY.json  ($(length(topo.gnb_locations)) sites)")
 
-open(joinpath(outdir, "meta.json"), "w") do f
-    print(f, "{\"country\":\"spain\",\"agents\":$NAG,\"duration\":$(round(Int,DURATION)),",
+open(joinpath(outdir, "meta-$COUNTRY.json"), "w") do f
+    print(f, "{\"country\":\"$COUNTRY\",\"agents\":$NAG,\"duration\":$(round(Int,DURATION)),",
             "\"dt\":$(round(Int,DT)),\"speed_kmh\":$SPEED,\"nsteps\":$NSTEPS,",
-            "\"edge_upfs\":52,\"psas\":5,\"handovers\":$nho}")
+            "\"edge_upfs\":$NEDGE,\"psas\":$NUM_PSA,\"handovers\":$nho}")
 end
 println("Done. Output in $outdir")

run_national.jl

@@ -21,18 +21,23 @@ const SCALE = 1000
 const NUM_PSA = 5                 # centralized PSAs (config.toml num_centralized_upfs)
 const ADOPTION = 0.82             # mobile_adoption_rate (config.toml)
 
-# (data subdir, mcc csv files, OpenCellID operator net id, #edge UPFs, population)
+# (data subdir, gNB csv files relative to data/<sub>, operator net id, #edge UPFs, population)
+#   usa     = OpenCellID Verizon (operator-tagged, sparse — lower density bound)
+#   usa_asr = FCC ASR all macro structures (operator-agnostic, complete — upper bound;
+#             net=999 synthetic tag so no operator filter applies). Density-invariance
+#             cross-check: same σ advantage on both real datasets.
 const PROFILES = Dict(
-    "spain" => ("spain", ["214.csv"],            7,   52,  49_442_844),
-    "usa"   => ("usa",   ["310.csv","311.csv"], 480, 817, 335_000_000),
+    "spain"   => ("spain", ["opencellid/214.csv"],                  7,   52,  49_442_844),
+    "usa"     => ("usa",   ["opencellid/310.csv","opencellid/311.csv"], 480, 817, 335_000_000),
+    "usa_asr" => ("usa",   ["asr/310.csv"],                        999, 817, 335_000_000),
 )
 
 function build_topology()
     haskey(PROFILES, COUNTRY) || error("unknown country $COUNTRY")
     sub, files, opid, nedge, _pop = PROFILES[COUNTRY]
-    base = joinpath(@__DIR__, "data", sub, "opencellid")
+    base = joinpath(@__DIR__, "data", sub)
     paths = filter(isfile, [joinpath(base, f) for f in files])
-    isempty(paths) && error("no OpenCellID data under $base")
+    isempty(paths) && error("no gNB data under $base for $files")
     # two_tier: nedge edge UPFs (UL-CL) clustered under NUM_PSA centralized PSAs
     cfg = SimConfig(1, 2, SCALE, 1, 1, 1, :two_tier, NUM_PSA, 1)
     topo = DSim.load_and_deploy_network(paths, opid, nedge, joinpath(@__DIR__, "data", sub), cfg)