一键导入
Measure system performance of this platform using
npx skills add https://github.com/AI-native-Systems-Research/ai-native-storage-certus --skill profile-performance-certus-api-bench复制此命令并粘贴到 Claude Code 中以安装该技能
Measure system performance of this platform using
npx skills add https://github.com/AI-native-Systems-Research/ai-native-storage-certus --skill profile-performance-certus-api-bench复制此命令并粘贴到 Claude Code 中以安装该技能
Add Coq 8.20 as an additional prover to an existing Creusot installation, for verification conditions that SMT solvers cannot discharge
Install the Creusot Rust verification tool
Synchronize Spin/Promela models with current source code and re-verify
Prepare an existing Cargo project to use Creusot formal verification
Create a new Spin/Promela formal verification model for a system property
Create a new version of a component
| name | profile-performance-certus-api-bench |
| description | Measure system performance of this platform using |
| argument-hint | [output-dir] |
The purpose of this skill is to capture a snapshot of performance by running certus-server (release build) and certus-api-bench.py.
Create the output directory from the output-dir argument, for example: mkdir -p Write PROFILE.md to /PROFILE.md.
Collect data for combinations of the following parameters:
lspci -d <vendor>:<device> or equivalent, sort PCI addresses lexicographically,
and use the first D addresses.
Pass each selected device as repeated flags:
--device-pci --device-pci ... --device-pci
If D > N, skip that combination and record SKIPPED: requested D > available N in PROFILE.md.--memory-tier-size <bytes> (use bytes, e.g. 4294967296).certus-api-bench --clients <count>.certus-api-bench --block-size <bytes> (use bytes, e.g. 1048576).certus-api-bench --batch-size <count>.certus-api-bench --num-objects <count>.certus-api-bench --iterations <count>.Other requirements:
--format.--device-pci flags, --memory-tier-size, and --format.
c. Retry startup up to 3 times with a 5s delay if it fails. If startup still fails, record the combination as FAILED in PROFILE.md with stderr and continue.
d. After a healthy start, run a 1-iteration warm-up of certus-api-bench before measurement.
e. Run the measured iterations.
f. If certus-api-bench exits non-zero, capture stdout/stderr to <output-dir>/logs/<run-name>.log, mark the run as ERROR in PROFILE.md, retry once, and if still failing continue to the next combination.
g. Stop certus-server and collect logs.Run combinatorial experiments and build a result table in /PROFILE.md.
PROFILE.md must include a table with columns: device_count | pci_list | memory_tier_size_bytes | clients | block_size_bytes | batch_size | num_objects | iterations | throughput_GB_per_sec (populate) | throughput_GB_per_sec (hot lookup) | throughput_GB_per_sec (cold lookup) | notes
Include raw output file links and log links in PROFILE.md.
Summarize the data and identify performance anomalies.
Create a plot with Y-axis throughput, X-axis device_count for:
Save plot in output directory.