Benchmark Performance Results

This section details the performance characteristics of the Altius execution engine across two distinct environments: Raw Execution (theoretical maximum) and End-to-End (real-world application).

Raw Execution Micro-Benchmarks (revm)

These metrics isolate the execution engine to measure maximum theoretical throughput for DeFi-heavy workloads using a dataset of 1,000 Uniswap transactions across 8 threads at a 0% conflict rate.

Metric	Sequential	Parallel + OCCDA	Parallel + SSA	Optimistic
Execution Time	38.3 ms	5.89 ms	6.89 ms	3.01 ms
Total Time	40.73 ms	12.08 ms	12.1 ms	8.28 ms
TPS	~24.5k	~48.8k	~82.6k	~120.7k
GPS (Mgas/s)	3,365	11,344	11,330	16,550
Exec Speedup	1.0x	6.5x	5.56x	12.72x
Conflict %	0%	0%	0%	0%

The optimistic strategy demonstrates a 12.72x speedup over standard sequential execution, achieving near-linear scaling and processing over 16.5 Gigagas per second in isolation.

End-to-End Node Benchmarks

These tests evaluate performance in a realistic OP L2 node environment, measuring the full block processing pipeline including database I/O and state root calculation.

Scenario A (Zero Conflict): In an ideal state with no transaction dependencies, parallel execution provides a 4.15x speedup in end-to-end latency.

Strategy	Server E2E Time	Client Latency	GPS (Ggas/s)	Speedup
Sequential	148.04 ms	440.92 ms	0.69	1.00x
Parallel	43.57 ms	47.26 ms	2.86	4.14x
Optimistic	43.46 ms	46.95 ms	2.82	4.15x

Scenario B (High Contention): With 50% of transactions conflicting, the engine maintains a 1.93x speedup over sequential processing.

Strategy	Server E2E Time	GPS (Ggas/s)	Speedup
Sequential	114.13 ms	1.09	1.00x
Parallel	76.94 ms	1.61	1.55x
SSA	89.99 ms	1.38	1.30x
Optimistic	64.58 ms	1.87	1.93x

The optimistic scheduler maintains superior performance even as conflicts rise, degrading gracefully while ensuring client latency remains stable. In realistic end-to-end scenarios, the system sustains between 1.8 and 2.8 Gigagas per second.

These results demonstrate that the Altius architecture successfully shifts the bottleneck away from the CPU, allowing for a 2–4x performance boost on standard node infrastructure.