Quick Reference Card
All formulas, thresholds, multipliers, and decision tables. Print this page for design reviews.
Step 1 U_eff = (H × 1.0) + (AI × 3.5) + (IoT × 0.2) + (Pod × 7.0)
Step 2 AIW = Σ(stream_Mbps) × Burst_factor
Step 3 CS = PII + Regulatory + Exposure + Model_sensitivity + Financial
(each component 0, 0.5, or 1.0)
Step 4 LL (1–5) → RTT budget:
LL1=500ms LL2=200ms LL3=80ms LL4=31ms LL5=20ms
RTT = 2×(Access+Core+FW+WAN+LB) + Inference + Jitter
Headroom = LL_budget − RTT [must be > 0]
Step 5 B_min = (U_eff × AIW × Peak_factor) / Utilisation_target
B_rec = B_min × 1.5
Model_sync_BW = GB × Updates/day × 8000 / 86400
B_total = B_rec + Model_sync_BW
Step 6 IS = (U_eff × AIW × CS × LL) / (B × A)
Step 7 PUO = IS / U_eff
Step 8 Route = f(LL, CS, Data_residency)
LL=5 → Tier 4 on-prem GPU
LL=4 + CS≥4 → Tier 3 campus edge
LL=4 + CS≤3 → Tier 3 campus edge
LL=3 + CS≤4 → Tier 2 regional hub
LL≤2 + CS≤3 → Tier 1 cloud
Any LL + CS=5 → Tier 3 or Tier 4 (compliance override)
Step 9 B_req = (U_eff × AIW × CS × LL) / (3.0 × A)
B_gap = B_req − B_current
Entity multipliers
| Entity |
Multiplier |
| Human operator |
× 1.0 |
| AI software agent / bot |
× 3.5 |
| IoT / edge sensor |
× 0.2 |
| GPU inference pod |
× 7.0 |
IS verdict
| IS |
Action |
| ≤ 1 |
Optimal — monitor quarterly |
| 1–3 |
Monitor — enforce QoS |
| 3–10 |
Upgrade required — 90-day plan |
| > 10 |
Blocker — no AI go-live |
PUO verdict
| PUO |
Action |
| < 0.3 |
Over-provisioned |
| 0.3–1.0 |
Balanced |
| 1.0–2.0 |
Stressed |
| > 2.0 |
Overloaded |
CS overhead
| CS |
Overhead |
| 1 |
+8% |
| 2 |
+18% |
| 3 |
+28% |
| 4 |
+40% |
| 5 |
+60% |
A factor
| Infrastructure |
A |
| Dark fiber / leased line |
0.95–1.00 |
| Dedicated MPLS SLA |
0.90–0.95 |
| SD-WAN dual-path |
0.82–0.90 |
| Shared MPLS |
0.70–0.82 |
| SD-WAN shared internet |
0.65–0.75 |
| Single ISP legacy |
0.50–0.65 |
AIW reference
| Workload |
Mbps |
Burst |
| STT |
0.3–0.5 |
1.5 |
| LLM text |
0.5–2.0 |
1.3 |
| Agent assist |
2.0–4.0 |
1.6 |
| Full AI agent |
5.0–8.0 |
1.8 |
| Vision API |
5–25 |
2.0 |
| Video analytics |
8–50 |
2.2 |
| RAG |
1–5 |
1.4 |
QoS 6-class model
| Class |
DSCP |
Workload |
| Real-time AI |
EF (46) |
STT, fraud, voice AI |
| Agent assist |
CS5 (40) |
LLM, coaching |
| RAG / knowledge |
CS4 (32) |
Vector DB |
| Analytics |
CS3 (24) |
Sentiment, batch |
| Model sync |
CS1 (8) |
Off-peak only |
| Background |
BE (0) |
Everything else |
ARS threshold
ARS ≥ 70 required before AI production go-live.
| Domain |
Max |
| Bandwidth capacity |
20 |
| Latency & switching |
20 |
| Security & compliance |
20 |
| Resilience & HA |
20 |
| Observability |
20 |
| Total |
100 |
Key design rules
- Never run AI links above 70% utilisation
- Size each redundant path for 100% of B_recommended (not 50%)
- LL = 4 mandates campus-edge AI — cloud is physically impossible
- LL = 5 mandates on-prem GPU — zero WAN
- IS > 10 = no AI go-live, full stop
- ARS < 70 = pilot only, not production
- Model sync must be off-peak — schedule 02:00–06:00
- PUO > 2.0 = agents will abandon AI tools within weeks
- Growth buffer: B_recommended × 1.5 minimum
- N+1 GPU pods: always one more pod than the minimum required
B point reference — where to calculate IS
| B point |
Layer |
Typical B |
A factor |
Usually the bottleneck? |
| B1 |
Agent access port |
1G |
1.00 |
No — 0.7% utilisation |
| B2 |
Access switch uplink (10G) |
10G |
0.97 |
No |
| B2* |
Access switch uplink (legacy 1G) |
1G |
0.95 |
YES — IS = 3.54 |
| B3 |
Core to SD-WAN handoff |
100M–1G |
0.70 |
YES — most common branch failure |
| B4 |
WAN internet circuit |
varies |
0.70 |
YES — without edge AI |
| B5 |
MPLS 10G WAN |
10G |
0.92 |
YES — without MCP tiering |
| B6 |
Internet 5G ISP |
5G |
0.72 |
YES — without MCP tiering |
| B7 |
5G cellular backup |
300–600M |
0.60 |
Emergency only |
| B8 |
Campus access uplink (25G) |
25G |
0.97 |
No with dual 25G PC |
| B9 |
Core to in-house DC |
100G |
0.99 |
Never |
Complete IS matrix — branch and campus
| Site |
Layer |
IS (naive) |
IS (post-MCP) |
Action |
| Branch |
B1 — access port |
0.007 |
0.007 |
No action |
| Branch |
B2 — 10G uplink |
0.347 |
0.347 |
No action |
| Branch |
B2* — legacy 1G |
3.54 |
3.54 |
Upgrade to 10G now |
| Branch |
B3 — 100M circuit |
96.2 |
0.94 |
Branch edge AI + 1G circuit |
| Branch |
B4 — 1G internet |
9.60 |
0.94 |
MCP tiering + 1G |
| Campus |
B5 — MPLS 10G |
9.77 |
0.32 |
MCP tiering — no BW upgrade |
| Campus |
B6 — Internet |
24.9 |
0.82 |
MCP tiering — no BW upgrade |
| Campus |
B7 — 5G backup |
— |
2.47 |
Emergency use only |
| Campus |
B8 — dual 25G PC |
0.75 |
0.75 |
No action |
| Campus |
B9 — DC 100G |
0.91 |
0.91 |
No action |