This repository provides public releases of Ant Group's production serverless FaaS traces, made accessible to the research and academic community.
These files include every cold start recorded in our public cloud.
Duration: 1-day period (complete data will be accessible prior to the conference).
| Metric/link | Description |
|---|---|
| 20250420.csv | 1-day period |
Schema
This table presents individual cold start events, with each row containing trace information for a cold start event. The details include the timestamp, cluster name, function name, request ID, pod ID, memory cost, total duration, time taken to allocate a pod, time to deploy code, and time to create an instance.
| Name | Description | Unit | Example |
|---|---|---|---|
| day | day | date | |
| time | timestamp | date | |
| clustername | cluster name | - | |
| funcname | MD5-hashed function name | - | |
| requestid | MD5-hashed traceId | - | |
| podid | MD5-hashed instanceId | - | |
| mem | instance memory cost | byte | |
| totalcost | total time spent on cold start | ms | |
| podallocationcost | schedule overhead | ms | |
| deploycodecost | time to download, extract, and deploy function code | ms | |
| createinstancecost | Function instance cold start time | ms |