# Serverless Performance Explorer

> Lambda REPORT lines → cold-start rate, p50/p95/p99, memory right-sizing and cost per 1M calls. Paste AWS Lambda REPORT log lines, JSON platform.report records or a CSV/Logs Insights export. Get cold-start rate, init and duration percentiles, memory utilisation with a right-sizing suggestion, and the cost per 1 million invocations on x86_64 and arm64 using the public Lambda prices (overridable).

Part of usefulweb.site (https://usefulweb.site/). Results are deterministic calculations from the inputs and the listed assumptions; the API returns exactly what the web page shows.

## Use it
- Human interface: https://serverlessperformanceexplorer.usefulweb.site/ (interactive tool at https://serverlessperformanceexplorer.usefulweb.site/tool)
- JSON API: POST https://serverlessperformanceexplorer.usefulweb.site/api/v1/calculate with a JSON object of inputs (schema: https://serverlessperformanceexplorer.usefulweb.site/api/v1/tool)
- OpenAPI: https://serverlessperformanceexplorer.usefulweb.site/openapi.json
- MCP (Streamable HTTP): https://serverlessperformanceexplorer.usefulweb.site/mcp - tool `serverless_performance_explorer`

## Inputs
- logs (textarea): REPORT lines or CSV - CloudWatch REPORT lines, JSON log-format platform.report records, or CSV with columns like requestId,duration,billedDuration,memorySize,maxMemoryUsed,initDuration[,function,timestamp]. Separate functions with "=== function: name ===".
- file (file): Or upload a log/CSV file
- architecture (select): Current architecture
- priceGbSecondX86 (number): Price per GB-second (x86_64) - Default: us-east-1 first tier, verified 2026-10-14.
- priceGbSecondArm (number): Price per GB-second (arm64)
- pricePerMillionRequests (number): Price per 1M requests
- monthlyInvocations (integer): Monthly invocations (optional) - Adds a monthly cost estimate (free tier not deducted).
- headroomPercent (number): Memory headroom - Added to the peak observed memory when suggesting a size.

## Method
- Each REPORT line (plain text or JSON "platform.report") yields Duration, Billed Duration, Memory Size, Max Memory Used and, on cold starts, Init Duration (or SnapStart Restore Duration). "Status: timeout|error" and "Task timed out after" lines mark failed invocations.
- Cold-start rate = invocations with Init/Restore Duration ÷ all invocations. Percentiles use the nearest-rank method on the observed values.
- GB-seconds per invocation = Billed Duration (s) × Memory Size (MB) ÷ 1024. If Billed Duration is missing it is estimated as ceil(Duration + Init) because INIT is billed for on-demand functions since 1 Aug 2025.
- Cost per 1M invocations = price per 1M requests + 1,000,000 × mean GB-s × price per GB-s, for x86_64 and arm64 (first pricing tier, no free tier).
- Right-sizing: suggested memory = peak Max Memory Used × (1 + headroom), rounded up to 64 MB (128–10,240 MB). Increase if peak use ≥ 90 % of the configured size; decrease if the suggestion is ≤ 75 % of it; otherwise keep. Cost estimates assume the billed duration does not change with memory (CPU scales with memory, so verify).
- CloudWatch Logs Insights exports report @memorySize/@maxMemoryUsed in bytes; values above 100,000 are converted to MB (10^6 bytes).