The Signaling Cost Model¶

The simulator does not measure signaling. It counts mobility events and charges each one a constant taken from the relevant 3GPP procedure or 6G-RUPA operation. This page gives every constant, its decomposition, and the clause it comes from, so that a reader can recompute or contest any of them without rerunning anything.

That split is deliberate and is what makes the results checkable: simulation supplies the event mix, analysis supplies the per-event cost, and the two never touch.

The procedure behind each constant on this page is drawn out message by message in Handover Sequence Diagrams, which is where to check a byte count against the message set it came from.

What \(\sigma\) Is¶

\(\sigma\) is the transient control-plane traffic exchanged to complete one mobility event, in bytes per event. It is not:

• steady-state memory, which is forwarding-table occupancy rather than messages;
• core forwarding-state churn \(\Delta S_{\mathrm{core}}\), a separate per-event quantity counting per-session writes;
• the radio leg. RRC reconfiguration and random access are identical under both architectures over the same NG-RAN, so they are excluded from both sides. Their inclusion would add the same constant to numerator and denominator.

All values are information-element-level estimates from the named clause, not packet captures. The headline is a ratio, so it survives any uniform rescaling of the set.

Where the Charge Is Decided¶

flowchart TD
    A[agent moves one tick] --> B{serving cell changed?}
    B -- no --> A
    B -- yes --> C{same edge UPF?}
    C -- yes --> D["d = 1<br/>access tunnel endpoint"]
    C -- no --> E{same session anchor?}
    E -- yes --> F["d = 2<br/>serving edge UPF relocates"]
    E -- no --> G{SSC mode}
    G -- "mode 1, anchor pinned" --> H["d = 2 charge<br/>plus path-length excess"]
    G -- "mode 2 or 3" --> I["d = 3<br/>anchor relocation"]
    D --> J["5G: 600 B<br/>6G-RUPA: 200 B"]
    F --> K["5G: 1150 B<br/>6G-RUPA: 200 B"]
    H --> K
    I --> L["5G: 2200-2700 B<br/>6G-RUPA: 200 B"]

The same event stream feeds both columns. Only the charge differs, which is why the ratio is insensitive to the mobility model while the absolute totals are not.

5G: A Procedure Family That Grows With Reach¶

Each depth rewrites a different amount of per-session tunnel state, so each has its own procedure and its own cost.

\(d=1\), access tunnel endpoint, 600 B¶

The user moves between base stations served by the same edge UPF. Only the N3 tunnel endpoint is updated, through one PFCP Session Modification. Procedure: TS 23.502 Sec. 4.9.1.2.2.

\(d=2\), serving edge UPF relocation, 1150 B¶

The user crosses into another edge UPF's region. The uplink classifier changes while the anchor and the UE's IP address are preserved, so the old session must be released, a new one established, and the path repointed at both ends. Procedure: TS 23.502 Sec. 4.9.1.3.2 to 3.3.

\(d=3\), anchor relocation, 2200 to 2700 B¶

A new anchor is established, the old one released, and the user gets a new address. Under SSC mode 1 this never fires on geometric mobility, because the anchor is pinned; it is a deliberate re-anchor. The simulator therefore does not charge it per anchor-region crossing. Crossing a pinned anchor's region boundary costs path length, not signaling, which is what \(\rho\) measures. Procedure: TS 23.502 Sec. 4.3.5, TS 23.501 Sec. 5.6.9.

\(d=4\), roaming entry, 3250 B¶

Home-Routed entry anchors the visited session back to the home network. Decomposition: session-management create over N16 (600 B), N4 establishment in both networks (700 B), N9 path setup (700 B), N32-f security envelopes (800 B), N2 and NAS accept leg (450 B). Roaming registration is excluded, which is conservative in 5G's favour. Architecture: TS 23.501 Sec. 4.2.4; transaction: TS 23.502 Sec. 4.3.2.2.2; security: TS 33.501 Sec. 13; deployment: GSMA NG.113 Sec. 3.1.2 and 5.1.2 to 5.1.3.

6G-RUPA: One Operation at Every Depth¶

A renumbering is not a handover procedure. It is the architecture's native assign-a-new-address primitive, and it is the same operation at every depth.

sequenceDiagram
    participant N as moving node
    participant L as local neighbourhood
    participant P as flow peers
    N->>N: obtain synonym under the destination aggregate (50 B)
    N->>L: advertise the new address locally (50 B)
    N->>P: rebinding update, one per active flow (150 B each)
    Note over N,P: old address retained during the policy timeout
    Note over N,L: core prefix set unchanged, so no core write

Two properties follow, and both are load-bearing:

• No term depends on depth. The advertisement reaches only the neighbourhood the node moved into, because routing to the old and the new address coincides everywhere upstream of it. So the same 200 B applies at \(d=1\), \(d=2\) and the anchor-relocation equivalent alike.
• No term installs core state. The destination aggregate is fixed by topology and already advertised; the node takes an address under it. This is why \(\Delta S_{\mathrm{core}} = 0\) rather than merely small.

First entry into a layer adds one enrollment exchange, about 450 B in total, charged once and not per event.

The per-flow term¶

\(\sigma_{\mathrm{RUPA}} = 50 + 150F\) for a node carrying \(F\) active flows. Flatness in depth is unaffected, but the comparison against 5G is not flat in \(F\): 5G rewrites one tunnel per session however many flows it carries. At \(F \ge 4\) a renumbering exceeds the 600 B depth-one procedure, and at \(F \ge 8\) the 1150 B depth-two one. The scenarios here assume the one to two active sessions per user of the modelled eMBB profile. A device holding many concurrent flows inverts the signaling comparison, though neither \(\Delta S_{\mathrm{core}}\) nor \(\rho\) depends on \(F\).

From Per-Event Cost to Aggregate Advantage¶

Under SSC mode 1 only the first two depths occur, so a deployment's whole event mix reduces to one number: \(\beta\), the share of events that reach the serving edge UPF.

Read it as the fraction of control bytes 6G-RUPA does not send. The denominator is what an average handover costs 5G on a deployment whose mix is \(\beta\); the numerator is what the same handover costs 6G-RUPA, which is the same everywhere.

Because \(\bar{A}\) is monotone in \(\beta\) and \(\beta \in [0,1]\) by definition, the advantage is bounded in \([66.7\,\%, 82.6\,\%]\) for any deployment whatsoever. The sweep over deployments is therefore exhaustive by construction rather than by sampling, and a single measured topology landing outside the band would refute the model.

flowchart LR
    A[national field] --> B[k-means region partition]
    B --> C[mobility model]
    C --> D[event stream]
    D --> E["classify by depth"]
    E --> F["beta = share at d=2"]
    F --> G["Abar(beta), closed form"]
    D --> H["charge both architectures"]
    H --> I["measured advantage"]
    G -.->|agree to 0.006 pp<br/>over 81 runs| I

Across the 81 runs of results/national-sweep.csv, covering 27 operator fields in six countries, substituting each run's measured \(\beta\) into the closed form reproduces its simulated advantage to within 0.006 percentage points, median 0.003. The simulator contributes one number per run; the rest is arithmetic.

What Would Change These Numbers¶

• Different byte estimates. The constants are IE-level estimates. Since the reported quantity is a ratio, rescaling the whole set uniformly changes nothing; changing their relative sizes moves the band.
• Retaining the 5G control plane. With PFCP mapped onto the 6G-RUPA transport through a shim, the state and continuity results hold but \(\sigma\) stays 5G's. Every \(\sigma\) result must be labelled with the control plane it assumes.
• Many concurrent flows per device, as above.

Where This Lives in the Code¶

Spec Ledger¶

Cite specs by official clause, never by line number in any local copy.