performance.md

Performance Baseline — Order Execution Cycle

Baseline numbers for Soroban CPU instruction cost and memory usage across the three core order execution scenarios. Generated by running:

cargo bench -p order-handler --bench order_execution

Note: Benchmarks are excluded from CI (cargo test) because they are too slow for routine checks. Run them locally or in a dedicated perf job with cargo bench -p order-handler --bench order_execution.

Soroban Budget Metrics

Scenario	CPU Instructions	Memory (bytes)	DS Reads	DS Writes	XC Calls
create_order	—	—	2	4	3
execute_order	—	—	8	6	5
full_cycle	—	—	18	14	10

Fill in CPU and memory columns by running the bench and reading the [bench] <scenario> cpu=… mem=… lines printed to stderr.

Metric Definitions

Metric	How it is measured
CPU Instructions	env.cost_estimate().budget().cpu_instruction_cost() after a reset to default limits
Memory (bytes)	env.cost_estimate().budget().mem_bytes_cost()
DS Reads	manual count of DataStoreClient::get_* calls in the hot path
DS Writes	manual count of DataStoreClient::set_* / add_* calls
XC Calls	number of cross-contract invocations (oracle, vault, role_store)

Optimisation Notes

• Issue #242 — validate_position was inlined into libs/position_utils (called as a library function from increase_position_utils / decrease_position_utils), eliminating at least one cross-contract round-trip per order execution.
• Issue #245 — All fee-computation paths use the named TOKEN_PRECISION constant from libs/math; no bare 10_000_000 literals remain.
• Future work tracked in issues #239–#242.