Stream Ingest API Performance Benchmark
Following are end to end latency results observed from benchmarking the Stream Ingest API. Note that latency here is the measurement of how long the Stream Ingest API call takes to complete. Since the API call is synchronous and blocking, features ingested are available for online serving immediately.
Setup​
- Number of PushSource(s): 1
- Number of Feature Views: 1
- Write to Online Store: True
- Write to Offline Store: True
- Includes Transformation: True
- Online Store: Dynamo
Note : All Latencies below are in milliseconds.
Batch Size : 1​
| 100 QPS | 500 QPS | 1000 QPS | |
|---|---|---|---|
| p50 | 57 | 55 | 55 |
| p90 | 71 | 69 | 67 |
| p95 | 86 | 81 | 76 |
| p99 | 165 | 127 | 108 |
Batch Size : 10​
| 20 QPS | 100 QPS | 200 QPS | |
|---|---|---|---|
| p50 | 98 | 112 | 82 |
| p90 | 115 | 135 | 97 |
| p95 | 124 | 151 | 143 |
| p99 | 156 | 213 | 190 |