Load Sharing

1) Why "joint" distribution

• stabilizes SLO during spikes and local failures;
• Reduces cost-to-serve
• increases fairness and predictability for roles;
• minimizes "noisy neighbors" and inter-domain arbitration.

2) Objects and roles

Power providers: validators/nodes, sequencers, DA pools, GPU/CPU clusters, POP/edge.
Consumers: service operators, creators/studios, affiliates/aggregators, analytics/ML.
Coordinators: balancers, routers, Policy/Compliance Gate, Rewards & Billing.
Supervision: auditory/regulyatory,治理 Committee.

3) Taxonomy of loads (QoS classes)

Q4 - deadline commands: critical order/finality (bridges, payments, risk).
Q3 - ordered streams: causality by key (user/session/asset).
Q2 - exactly-once effective: billing/snapshots/rights transfer.
Q1/Q0 - mass/best effect: telemetry, indexes, offline analytics.

For each class, SLO/SLA, retray windows, in-flight limits, priorities are recorded.

4) NRC policies: what we optimize

Utility(route    provider) =
wL·Latency_p95 + wQ·QueueDepth + wC·Cost_per_unit
+ wF·FinalityLag  + wR·RiskScore + wA·AvailabilityPenalty
+ wG·Geo/PolicyPenalty

• Q4 ↑wL, ↑wF, ↑wR; Q1 ↑wC, ↓wF.

Invariants: Strict-order per key (Q3/Q4), idempotency, RNFT/compliance limits.

5) Joint distribution algorithms

Consistent Hashing per key with Hot-Shard Relief (temporary hot key subsegmentation).
Percentile-aware routing: p95/p99 solution, not p50, so as not to hide tails.
Capacity-aware quotas: token-buckets per QoS class/provider/region.
EDF/LLF для Q4: Earliest Deadline First / Least Laxity First.
Probing & Half-open: quick samples of "recovery" of the derived routes.
Backpressure: shapers, max-in-flight, degradation by policy (graceful).
Dual-write/Replay barriers (Q3/Q2): for secure transfer between providers.

6) Justice and anti- "noisy neighbor"

• Jain Fairness Index по CPU/GPU/IO/egress; The target corridor is supported by quotas.
• Weighted fair queuing (WFQ/DRR) on public queues;
• Budget-limits on value and volume;
• Surge surcharges on congested directions (dynamic wC);
• Fines for systematic excess of tails/errors.

7) Economics and incentives

Charging units: vCPU-sec, GiB-hour RAM, GPU-minute, GB-storage-month, GB-egress, DA-byte.

[
P_i = \sum_t \underbrace{\text{Rate}i \cdot U{i,t}}{\text{объем}}
\ cdot\underbrace {QF {i, t}} {\text {quality}}
-\underbrace {Penalty {i, t}} _ {\text {SLA/incidents}}
]

where (QF) is the multiplier for SLO (success, p95, DLQ = 0, finality lag).

Quality bonus: domains with stable SLO receive the -rate ↓take or traffic ↑obyem.
Insurance fund/slashing: covers compensation; managed by S-pledges in RNFT.

8) RNFT contracts and rights

• `role_bindings` (Provider/Operator/Oracle/Sequencer), `shares/fees`, `QoS-классы`;
• `quotas/limits`, `S-stake`, `slashing_rules`, `SLA/KPI`;
• 'region/compliance '(whitelisting), 'egress/DA' ceilings;
• `dispute/escrow`, `governance_version`, `sunset`.

9) Order, idempotence, finality

Strict-order per key on the selected route; with failover - "pause" + replay barrier.
Outbox/Inbox + idempotency_key and seen-tables (TTL).
X-chain finality: accounting for challenge windows; critical operations are directed to the minimum'FinalityLag '.

10) Compliance and geo-rules

Fail-closed: in doubt - blocking, manual quorum.
ZK passes: age/geo/sanction verification without personal data disclosure.
Taxes/Deductions: In the way of payments through Rewards Router.
Data export policies: DA/egress by region, retention periods.

11) Observability and telemetry

End-to-end tracing: 'x _ msg _ id', 'route _ id', 'provider _ id', bridge/DA stages.
Metrics (per QoS/provider): p50/p95/p99, retry%, timeout%, duplicate ratio, out-of-order%, queue depth, finality lag, cost/req.
Дашборды: Shared Load Live, Tail Heatmap, Provider Quality, Cost-per-Route, Fairness Panel.
Alerts: error-budget burn, flap-rate, DLQ depth, surge prices, compliance blocks.

12) Incidents and degradation

1. Detection: p95/p99 growth, queues, finality lag, compliance errors.
2. Isolation: trip circuit, redistribution of shares, lowering quotas for noisy streams.
3. Compensation: payments from escrow/insurance fund under RNFT rules.
4. Post-mortem: RCA, update of weights/limits/risk signatures, rehearsal.

13) Formulas and landmarks

SuccessRate = 1 − (timeouts+errors)/requests

TailAmplification = p99/p50 (target: ↓, corridors per QoS)

FairnessIndex (Jain) = (Σ x) ²/( n· Σ x ²) by quota/resource

Cost/Req = Σ (resource × bid )/successful _ requests

Headroom = (cap − current)/cap

Provider QualityFactor: (QF = f (\text {success}, p95, DLQ, finality))

Utility_min при `Order=true ∧ Compliance=true ∧ Quotas=true`

• Q4: success ≥ 99. 99%, p95 ≤ 200 ms, DLQ = 0, MTTR ≤ 15 min.
• Q3: violation of the order of ≤ 10⁻⁶/soobshch., p95 ≤ 500 ms.
• DA: finality ≤ 3 × T _ block at Throughput ≥ X GB/h.

14) 治理 (weights, quotas, prices)

Proposals: change in weights (w), limits, tariffs and quality bonuses.
R-modifier: Votes in quorum quality are weighted by R reputation.
Sunset edits: temporary changes → auto-rollback without reselection.
Public reporting: Quarterly reports on provider quality and equity.

15) Implementation playbook

1. Mapping flows and causality keys (by QoS/region/compliance).
2. Definition of providers and their RNFT frames (quotas, pledges S, KPI).
3. Telemetry and samples (OWD/RTT/jitter/queue/cost/finality; EWMA+p95/p99).
4. Utility policies (weights per QoS, cost budget, surge corridors).
5. Delivery guarantees (outbox/inbox, idempotence, serial barriers).
6. Backpressure and fairness (WFQ/DRR, token buckets, anti-noise).
7. Observability (dashboards, alerts, error budgets).
8. Chaos/game-days (provider/bridge/DA drop, bursts, geo-blocks).
9. Economy and reverse (QF bonuses, fines/slashing, escrow).
10. 治理 and reporting (proposals, sunset, public metrics).
11. Scaling (new providers/regions, route optimization).

16) NRC Program KPI

Delivery: success (per QoS), DLQ = 0 (Q4/Q3), duplicate/out-of-order ↓.
Delay: p95/p99 and TailAmplification in target corridors.

Fairness: Jain ≥ target, reducing incidents "noisy neighbor."

Economy: Cost/Req ↓ with constant SLO, an increase in the share of "cheap" routes.
Robustness: MTTR median ≤ target, stable flap-rate.
Compliance: 100% pass geo/age/sanctions, zero violations.
Providers: the share of volume from providers with a high QF ↑, the frequency of fines ↓.

17) Delivery checklist

• QoS classes, causality keys and SLO/SLAs defined
• Per route/provider Utility Policies, Quotas and Tokens Configured
• Implemented consistent hashing, hot-shard relief, EDF/LLF for Q4
• Includes outbox/inbox, idempotency, and ordinal barriers
• Telemetry and dashboards connected (latency/tail/queue/cost/finality)
• Backpressure and fairness (WFQ/DRR, anti-noise)
• Set up QF bonuses/penalties, escrow and S-slashing
• Passed chaos/game-days and post-mortems
• Compliance Gate and tax deductions work
• Utverzhden治理 - Weights/Limits/Prices process (with sunset)

18) Glossary

NRC: cooperative load distribution.
RNFT: non-interchangeable relationship/entitlement/limit contract and KPI.
QF (Quality Factor): multiplier of payments/volume by provider quality.
Tail Amplification: p99/p50 - tail strength.
WFQ/DRR: a family of weighted equity planners.
Outbox/Inbox: Pattern of guaranteed delivery and idempotency.
Surge pricing: dynamic overload surcharge.

19) The bottom line

Shared load sharing turns the network into a cooperative processing pool, where politics (QoS, fairness, compliance) and economics (QF bonuses, fines, pledges) direct traffic to where it will be processed quickly, honestly and inexpensively - without losing order and finality. Such an outline provides predictable SLOs, transparent incentives for providers and resilience to peaks, crashes and price shocks.