Summary
Current benchmarks cover basic cold-start and density. Align with the benchmarking approach used in hyperlight-dev/hyperlight (core) and the dedicated benchmarks repo to provide more comprehensive and comparable numbers.
Areas to investigate
- What metrics does hyperlight-core track that we should also track? (e.g., guest call latency, host function dispatch overhead, memory per VM)
- Can we reuse or adapt the benchmarks repo infrastructure?
- Add warm-start (snapshot/restore) latency benchmarks — especially relevant for pyhl
- Add throughput benchmarks (requests/sec for networking workloads)
- Ensure historical trends are comparable across releases
Summary
Current benchmarks cover basic cold-start and density. Align with the benchmarking approach used in hyperlight-dev/hyperlight (core) and the dedicated benchmarks repo to provide more comprehensive and comparable numbers.
Areas to investigate