Traffic benchmarks
1) Why do we need benchmarks
Traffic benchmarks are repeatable tests that allow you to compare network configurations, providers and microgrids in terms of speed, stability, cost and compliance with invariants (order, finality, compliance). Ones:- give a numerical base for SLO/SLA and QF bonuses;
- identify bottlenecks (tail, queues, finality, egress);
- allow fair matching of domains/chains/providers;
- serve as the input of dlya治理 solutions (quotas, prices, limits).
2) Scope
Transport/routing: POP/edge, anycast, QUIC/HTTP/3.
Messaging & Order: outbox/inbox, idempotency, strict-order per key.
DA & Бриджи: throughput/GB, finality lag, challenge windows.
Multi-chain: cross-chain events and their finality.
Economy: Cost/Req, Cost/GB DA, Cost/egress.
Compliance: geo/age/sanctions, data export policy.
Observability: trace, p95/p99 metrics, DLQ/replay.
3) QoS classes (load profiles)
Q4 - deadline teams: order and finality are critical.
Q3 - ordered streams: causality by key (user/session/asset).
Q2 - exactly-once effective: billing/snapshots/rights transfer.
Q1/Q0 - best effect: analytics/indexing/telemetry.
For each profile, the volume, traffic template, deadlines, allowed retrays are set.
4) Benchmark set (suite)
4. 1 Microtests (micro-SUT)
Latency RTT/OWD: p50/p95/p99, TailAmplification = p99/p50.
Throughput max: msgs/s, tx/s, GB/s (DA/egress).
Queue/Backpressure: depth of queues, convergence time.
Idempotency cost: duplicates, deduplication time.
Order violations: out-of- order/10⁶ messages.
4. 2 Macro tests (end-to-end)
Route Utility: comparison of routes by utility function.
Finality-aware: Time to finality on breeches/DA.
Cost-aware: request price for fixed SLOs.
Compliance-gated: speed/success of passing geo/age-rules.
4. 3 Resilience
Soak/Endurance: 24-72 h stable load.
Chaos/Game-Day: POP/bridge/DA drop, price shock, geo-block.
Hot-Shard Relief: bursts of hot keys, subsegmentation.
5) Methodology
5. 1 Traffic generators
By causality key: consistent hashing, adjustable key cardinality.
Load patterns: Poisson/Bursty/Trace-replay, "sports" for Q4.
Retrai: exponential backoff with jitter, limits per QoS.
5. 2 Environment
Topologies: mesh/hub-and-spoke/edge-centric.
Regionality: inter-POP delays, cross-continent.
Stabilization: warm-up, steady window, cool-down.
5. 3 Invariants (hard)
`Order=true` (per key для Q3/Q4), `Idempotency=true`, `Quotas=true`, `Compliance=true`.
6) Reference metrics and formulas
SuccessRate = 1 − (timeouts + errors)/requests
TailAmplification = p99/p50 (Target: ↓)
Headroom = (cap − current)/cap
Cost/Req = Σ (resource × bid )/successful _ requests
FairnessIndex (Jain) = (Σ x) ²/( n· Σ x ²) by quota/resource
FinalityScore = f(lag, variance, reorgs)
QualityFactor (QF) = f(success, p95, DLQ, finality)
SLO landmarks (example):- Q4: success ≥ 99. 99%, p95 ≤ 200 ms, DLQ = 0, MTTR ≤ 15 min.
- Q3: out-of-order ≤ 10⁻⁶/soobshch., p95 ≤ 500ms.
- DA/Bridge: finality ≤ 3 × T _ block, false confirmations = 0.
7) Benchmark matrix (example)
8) Utilitarian route function
Utility(route) =
wL·Latency_p95_EWMA
+ wJ·Jitter
+ wQ·QueueDepth
+ wC·Cost_per_unit (gas + DA + egress + compute)
+ wF·FinalityTime
+ wR·RiskScore
+ wA·AvailabilityPenalty
+ wG·Geo/PolicyPenaltyBalance profiles - by QoS. Decisions are made at p95/p99 (percentiles, not averages).
9) Reports and dashboards
Traffic Bench Overview: p50/p95/p99, TA, success, cost/req for benches.
Finality & Bridge: finality, reorg, challenge.
DA/Storage: throughput/GB, retention, proof availability.
Queues & Backpressure: depth, lag, recovery time.
Fairness & Quotas: Jain, quota usage, noisy-neighbor.
Economy Panel: resource prices, share of "expensive" routes.
Compliance View: blocks/fouls, TTL, export/retention.
10) Use of results
Quotas and limits: redistribution by providers and routes.
Pricing: wC dynamics, discounts for sustainable quality.
RNFT edits: update KPI/fines/escrow.
治理: scale/policy change props (with sunset windows).
Incentives: QF bonuses to domains/providers that withstand SLOs.
11) Launch playbook (step by step)
1. Define goals and QoS: which SLOs we check and where it hurts.
2. Collect trace data: basic traffic profiles, causality keys.
3. Prepare the environment: topology, warm-up, trace 'x _ msg _ id'.
4. Set the retray/idempotency policy and the finality window.
5. Run micro tests: RTT/OWD, throughput, queue, order.
6. Run macro tests: utility routing, finality, cost.
7. Soak/Chaos: endurance 24-72h failure/shock scenarios.
8. Collect the report: metrics, deviations, RCA, recommendations.
9. Accept resheniya治理: quotas, prices, QF, limits, sunset edits.
10. Repeat the cycle: after optimizations - regress bench.
12) Benchmarking Program KPI
Quality: p95/p99 ↓, TA ↓ with fixed throughput.
Delivery: success ↑, DLQ depth ↓, replay success ↑.
Finality: FinalityTime ↓, false confirmations = 0.
Economy: Cost/Req ↓, the share of "cheap" routes ↑.
Justice: Jain in the hallway, "noisy-neighbor" fall.
Persistence: MTTR median ↓, flap-rate ↓.
Compliance: 100% pass geo/age/sanctions.
13) Delivery checklist
- QoS profiles, SLOs, and error budgets are described
- Configured traffic generators (keys, patterns, retrays)
- Included outbox/inbox, idempotency, ordinal barriers
- Finality and DA/bridge configuration windows set
- pXX metrics tracing and collection active/queues/cost/finality
- Utility profiles and QoS weights defined
- Ready panels and export reports, control thresholds
- Soak/Chaos Plan and Post Mortem Protocol
- Protsedury治理 and sunset windows for changes
- Insurance pool/escrow and QF mechanics of compensation
14) Glossary
TailAmplification: p99/p50 - "tail strength."
FinalityTime: window to irreversibility (bridge/DA/circuit).
Utility-routing: path selection by aggregated utility.
RNFT: Relationship/Rights/Limits Contract and KPIs.
QF: Quality payment multiplier.
DLQ/Replay: Quarantine/Reprocessing.
Soak/Chaos: Long-term and failure resistance tests.
15) The bottom line
Traffic benchmarks are an operational quality standard. They turn discussions about "faster/cheaper/more reliable" into reproducible numbers and manageable solutions: where to change quotas and prices, who to encourage with QF bonuses, how to rebuild routes and what to fix in infrastructure. A regular benchmarking cycle keeps the ecosystem in the SLO green zone at a controlled cost and without compromises in order, finality and compliance.