# Appendix C: Performance Data

#### Appendix C: Performance Characteristics

This appendix provides detailed performance data from benchmarks and production deployments.

#### Consensus Latency Benchmarks

Measured on AWS instances (c5.2xlarge, 8 vCPU, 16GB RAM) with simulated network latency:

| Peer Count | Avg Latency | p50   | p95    | p99    | Configuration        |
| ---------- | ----------- | ----- | ------ | ------ | -------------------- |
| 4          | 18ms        | 16ms  | 25ms   | 32ms   | Latency-optimized    |
| 8          | 26ms        | 24ms  | 35ms   | 44ms   | Latency-optimized    |
| 16         | 52ms        | 48ms  | 68ms   | 89ms   | Latency-optimized    |
| 32         | 103ms       | 98ms  | 128ms  | 156ms  | Latency-optimized    |
| 8          | 857ms       | 820ms | 1100ms | 1350ms | Throughput-optimized |

Real-world Benchmarks:

| Application       | Peer Count | Measured Latency | Duration           | Events                      |
| ----------------- | ---------- | ---------------- | ------------------ | --------------------------- |
| Horse racing game | 8          | \~26ms           | Multiple sessions  | 10,000+ consensus events    |
| Unity Boss Room   | 4-6        | \~35ms           | Devcom demo        | 5,000+ consensus events     |
| Robotics pilot    | 5          | 50-75ms          | 10-minute sessions | 800+ coordination decisions |

#### Throughput Benchmarks

Transactions per second (TPS) varies by configuration:

**Latency-optimized** (small batches, fast consensus):

| Peers | TPS     |
| ----- | ------- |
| 4     | 145,000 |
| 8     | 280,000 |
| 16    | 380,000 |
| 32    | 420,000 |

**Throughput-optimized** (large batches, slower consensus):

| Peers | TPS       |
| ----- | --------- |
| 8     | 1,026,000 |
| 16    | 1,450,000 |
| 32    | 1,820,000 |

Real applications operate well below maximum capacity, trading throughput for latency as required.

#### Network Overhead

Bandwidth consumption per participant:

| Configuration       | Gossip Traffic | Relay Traffic (if needed) | Total Bandwidth |
| ------------------- | -------------- | ------------------------- | --------------- |
| 4 peers, low TPS    | \~50 KB/s      | \~35 KB/s                 | \~85 KB/s       |
| 8 peers, medium TPS | \~120 KB/s     | \~75 KB/s                 | \~195 KB/s      |
| 16 peers, high TPS  | \~250 KB/s     | \~150 KB/s                | \~400 KB/s      |

Most applications use tunneling for 25-65% of traffic. Peer-to-peer connections (no tunneling) reduce bandwidth costs proportionally.

#### Lattice Infrastructure Performance

Orchestrator response times:

* Discovery request → Resource Node assignment: 15-45ms (p95)
* Proof validation: 5-20ms per proof
* Reward calculation and distribution: 100-300ms per session

Resource Node availability:

* Overall network uptime: 99.2%
* Individual high-reputation nodes: 99.7%+
* Geographic coverage: <100ms latency to nearest node for 95% of global population

#### Fault Tolerance Validation

Byzantine fault injection tests:

| Test Scenario       | Peer Count | Byzantine Count | Consensus Result  | Latency Impact |
| ------------------- | ---------- | --------------- | ----------------- | -------------- |
| Conflicting events  | 8          | 2               | Correct consensus | +15% latency   |
| Delayed messages    | 16         | 5               | Correct consensus | +25% latency   |
| Message withholding | 32         | 10              | Correct consensus | +30% latency   |
| Coordinated attack  | 8          | 3 (>f)          | Consensus halted  | N/A (expected) |

Tests confirm that consensus continues correctly with up to f Byzantine participants, but halts (maintaining safety) when Byzantine count exceeds f.