To ensure that changes to Chr.Avro don’t introduce major performance regressions (and to highlight areas where performance could be improved), we maintain a set of rudimentary benchmarks to test serialization throughput.
Each benchmark runs five times, and we take the average time of those runs. These results were collected on a 2017 MacBook Pro (3.1 GHz Intel Core i5):
| Benchmark | Average time (ms) | ||
|---|---|---|---|
| Chr.Avro | Confluent.Apache.Avro | ||
| boolean | deserialization 10,000,000 iterations | 66.970 | 204.664 |
serialization 10,000,000 iterations | 113.672 | 1007.223 | |
| double | deserialization 10,000,000 iterations | 303.675 | 436.088 |
serialization 10,000,000 iterations | 375.856 | 1385.805 | |
| float | deserialization 10,000,000 iterations | 245.849 | 454.685 |
serialization 10,000,000 iterations | 318.858 | 1257.844 | |
| int | deserialization 10,000,000 iterations | 167.990 | 379.699 |
serialization 10,000,000 iterations | 253.854 | 1204.909 | |
| long | deserialization 10,000,000 iterations | 238.436 | 537.021 |
serialization 10,000,000 iterations | 379.155 | 1115.019 | |
| record | deserialization 1,000,000 iterations | 187.314 | n/a |
serialization 1,000,000 iterations | 669.371 | n/a | |
| record (generic) | deserialization 1,000,000 iterations | n/a | 769.342 |
serialization 1,000,000 iterations | n/a | 636.208 | |
| record (specific) | deserialization 1,000,000 iterations | n/a | 409.978 |
serialization 1,000,000 iterations | n/a | 592.802 | |
| string | deserialization 10,000,000 iterations | 953.264 | 945.990 |
serialization 10,000,000 iterations | 1895.166 | 2144.516 | |
In the future, we’re hoping to expand the benchmarks to test more complex situations as well as non-.NET libraries.