Benchmark Results
Live benchmark comparisons between Dekaf and Confluent.Kafka, automatically updated on every commit to main.
Last Updated: 2026-03-14 11:53 UTC
info
These benchmarks run on GitHub Actions (ubuntu-latest) using BenchmarkDotNet. Ratio < 1.0 means Dekaf is faster than Confluent.Kafka
Producer Benchmarks
Comparing Dekaf vs Confluent.Kafka for message production across different scenarios.
| Method | Categories | MessageSize | BatchSize | Mean | Error | StdDev | Ratio | RatioSD | Gen0 | Gen1 | Allocated | Alloc Ratio |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Confluent_ProduceBatch | BatchProduce | 100 | 100 | 5,931.01 μs | 71.565 μs | 47.336 μs | 1.00 | 0.01 | - | - | 106.94 KB | 1.00 |
| Dekaf_ProduceBatch | BatchProduce | 100 | 100 | 1,362.99 μs | 67.892 μs | 40.401 μs | 0.23 | 0.01 | - | - | 47.35 KB | 0.44 |
| Confluent_ProduceBatch | BatchProduce | 100 | 1000 | 7,282.88 μs | 70.299 μs | 46.498 μs | 1.00 | 0.01 | 31.2500 | 15.6250 | 1063.29 KB | 1.00 |
| Dekaf_ProduceBatch | BatchProduce | 100 | 1000 | 2,290.42 μs | 105.874 μs | 63.004 μs | 0.31 | 0.01 | 23.4375 | 7.8125 | 577.51 KB | 0.54 |
| Confluent_ProduceBatch | BatchProduce | 1000 | 100 | 6,543.76 μs | 17.821 μs | 9.321 μs | 1.00 | 0.00 | 7.8125 | - | 194.43 KB | 1.00 |
| Dekaf_ProduceBatch | BatchProduce | 1000 | 100 | 1,148.01 μs | 55.850 μs | 33.235 μs | 0.18 | 0.00 | - | - | 60.12 KB | 0.31 |
| Confluent_ProduceBatch | BatchProduce | 1000 | 1000 | 9,494.19 μs | 1,999.892 μs | 1,190.104 μs | 1.01 | 0.16 | 78.1250 | 31.2500 | 1938.46 KB | 1.00 |
| Dekaf_ProduceBatch | BatchProduce | 1000 | 1000 | 5,412.45 μs | 343.502 μs | 204.413 μs | 0.58 | 0.07 | 46.8750 | 15.6250 | 1425.73 KB | 0.74 |
| Confluent_FireAndForget | FireAndForget | 100 | 100 | 112.24 μs | 47.699 μs | 31.550 μs | 1.07 | 0.40 | 1.9531 | - | 49.08 KB | 1.00 |
| Dekaf_FireAndForget | FireAndForget | 100 | 100 | 62.42 μs | 8.124 μs | 4.835 μs | 0.60 | 0.16 | 0.2441 | - | 6.48 KB | 0.13 |
| Confluent_FireAndForget | FireAndForget | 100 | 1000 | 1,275.58 μs | 26.347 μs | 17.427 μs | 1.00 | 0.02 | 15.6250 | - | 421.63 KB | 1.00 |
| Dekaf_FireAndForget | FireAndForget | 100 | 1000 | 709.34 μs | 78.219 μs | 51.737 μs | 0.56 | 0.04 | - | - | 89.25 KB | 0.21 |
| Confluent_FireAndForget | FireAndForget | 1000 | 100 | NA | NA | NA | ? | ? | NA | NA | NA | ? |
| Dekaf_FireAndForget | FireAndForget | 1000 | 100 | 478.27 μs | 35.369 μs | 21.048 μs | ? | ? | 0.4883 | - | 21.08 KB | ? |
| Confluent_FireAndForget | FireAndForget | 1000 | 1000 | NA | NA | NA | ? | ? | NA | NA | NA | ? |
| Dekaf_FireAndForget | FireAndForget | 1000 | 1000 | 4,621.48 μs | 638.369 μs | 379.883 μs | ? | ? | 7.8125 | - | 212.99 KB | ? |
| Confluent_ProduceSingle | SingleProduce | 100 | 100 | 5,386.42 μs | 16.885 μs | 11.169 μs | 1.00 | 0.00 | - | - | 1.57 KB | 1.00 |
| Dekaf_ProduceSingle | SingleProduce | 100 | 100 | 1,241.39 μs | 58.540 μs | 34.836 μs | 0.23 | 0.01 | - | - | 5.03 KB | 3.21 |
| Confluent_ProduceSingle | SingleProduce | 100 | 1000 | 5,368.04 μs | 17.411 μs | 10.361 μs | 1.00 | 0.00 | - | - | 1.56 KB | 1.00 |
| Dekaf_ProduceSingle | SingleProduce | 100 | 1000 | 1,303.01 μs | 49.939 μs | 33.031 μs | 0.24 | 0.01 | - | - | 5.03 KB | 3.22 |
| Confluent_ProduceSingle | SingleProduce | 1000 | 100 | 5,369.71 μs | 16.119 μs | 9.592 μs | 1.00 | 0.00 | - | - | 2.43 KB | 1.00 |
| Dekaf_ProduceSingle | SingleProduce | 1000 | 100 | 1,315.10 μs | 50.045 μs | 33.102 μs | 0.24 | 0.01 | - | - | 5.03 KB | 2.07 |
| Confluent_ProduceSingle | SingleProduce | 1000 | 1000 | 5,368.15 μs | 9.121 μs | 5.428 μs | 1.00 | 0.00 | - | - | 2.43 KB | 1.00 |
| Dekaf_ProduceSingle | SingleProduce | 1000 | 1000 | 1,115.77 μs | 35.831 μs | 18.741 μs | 0.21 | 0.00 | - | - | 5.03 KB | 2.07 |
Benchmarks with issues: ProducerBenchmarks.Confluent_FireAndForget: Job-TNLLLI(IterationCount=10, RunStrategy=Throughput, WarmupCount=3) [MessageSize=1000, BatchSize=100] ProducerBenchmarks.Confluent_FireAndForget: Job-TNLLLI(IterationCount=10, RunStrategy=Throughput, WarmupCount=3) [MessageSize=1000, BatchSize=1000]
Consumer Benchmarks
Comparing Dekaf vs Confluent.Kafka for message consumption.
| Method | Categories | MessageCount | MessageSize | Mean | Error | StdDev | Ratio | Allocated | Alloc Ratio |
|---|---|---|---|---|---|---|---|---|---|
| Confluent_ConsumeAll | ConsumeAll | 100 | 100 | 3.168 s | 0.0038 s | 0.0006 s | 1.00 | 76592 B | 1.00 |
| Dekaf_ConsumeAll | ConsumeAll | 100 | 100 | 3.011 s | 0.0036 s | 0.0006 s | 0.95 | 136096 B | 1.78 |
| Confluent_ConsumeAll | ConsumeAll | 100 | 1000 | 3.165 s | 0.0050 s | 0.0008 s | 1.00 | 256592 B | 1.00 |
| Dekaf_ConsumeAll | ConsumeAll | 100 | 1000 | 3.010 s | 0.0041 s | 0.0011 s | 0.95 | 327856 B | 1.28 |
| Confluent_ConsumeAll | ConsumeAll | 1000 | 100 | 3.165 s | 0.0018 s | 0.0005 s | 1.00 | 616592 B | 1.00 |
| Dekaf_ConsumeAll | ConsumeAll | 1000 | 100 | 3.011 s | 0.0052 s | 0.0013 s | 0.95 | 472272 B | 0.77 |
| Confluent_ConsumeAll | ConsumeAll | 1000 | 1000 | 3.165 s | 0.0018 s | 0.0005 s | 1.00 | 2424608 B | 1.00 |
| Dekaf_ConsumeAll | ConsumeAll | 1000 | 1000 | 3.011 s | 0.0007 s | 0.0002 s | 0.95 | 2286600 B | 0.94 |
| Confluent_PollSingle | PollSingle | 100 | 100 | 3.157 s | 0.0037 s | 0.0010 s | 1.00 | - | NA |
| Dekaf_PollSingle | PollSingle | 100 | 100 | 3.006 s | 0.0021 s | 0.0005 s | 0.95 | 31784 B | NA |
| Confluent_PollSingle | PollSingle | 100 | 1000 | 3.157 s | 0.0019 s | 0.0003 s | 1.00 | - | NA |
| Dekaf_PollSingle | PollSingle | 100 | 1000 | 3.007 s | 0.0020 s | 0.0005 s | 0.95 | 34640 B | NA |
| Confluent_PollSingle | PollSingle | 1000 | 100 | 3.157 s | 0.0038 s | 0.0010 s | 1.00 | - | NA |
| Dekaf_PollSingle | PollSingle | 1000 | 100 | 3.006 s | 0.0013 s | 0.0003 s | 0.95 | 32616 B | NA |
| Confluent_PollSingle | PollSingle | 1000 | 1000 | 3.157 s | 0.0039 s | 0.0010 s | 1.00 | - | NA |
| Dekaf_PollSingle | PollSingle | 1000 | 1000 | 3.006 s | 0.0017 s | 0.0003 s | 0.95 | 35792 B | NA |
Protocol Benchmarks
Zero-allocation wire protocol serialization/deserialization.
tip
Allocated = - means zero heap allocations - the goal of Dekaf's design!
| Method | Mean | Error | StdDev | Median | Allocated |
|---|---|---|---|---|---|
| 'Write 1000 Int32s' | 20.987 μs | 9.4948 μs | 5.6502 μs | 22.072 μs | - |
| 'Write 100 Strings (100 chars)' | 13.422 μs | 0.3211 μs | 0.2124 μs | 13.450 μs | - |
| 'Write 100 CompactStrings' | 11.317 μs | 0.1929 μs | 0.1276 μs | 11.322 μs | - |
| 'Write 1000 VarInts' | 34.771 μs | 12.7017 μs | 7.5586 μs | 37.374 μs | - |
| 'Read 1000 Int32s' | 17.396 μs | 14.4728 μs | 8.6126 μs | 22.151 μs | - |
| 'Read 1000 VarInts' | 29.935 μs | 14.3853 μs | 8.5604 μs | 34.084 μs | - |
| 'Write RecordBatch (10 records)' | 21.088 μs | 2.2757 μs | 1.3542 μs | 21.429 μs | - |
| 'Read RecordBatch (10 records)' | 4.321 μs | 0.2113 μs | 0.1397 μs | 4.319 μs | - |
Serializer Benchmarks
| Method | Mean | Error | StdDev | Ratio | RatioSD | Allocated | Alloc Ratio |
|---|---|---|---|---|---|---|---|
| 'Serialize String (10 chars)' | 1,496.7 ns | 229.12 ns | 136.35 ns | 0.36 | 0.03 | - | NA |
| 'Serialize String (100 chars)' | 1,634.8 ns | 126.99 ns | 83.99 ns | 0.40 | 0.02 | - | NA |
| 'Serialize String (1000 chars)' | 1,577.2 ns | 37.96 ns | 19.85 ns | 0.38 | 0.01 | - | NA |
| 'Deserialize String' | 2,937.0 ns | 97.86 ns | 58.23 ns | 0.71 | 0.02 | - | NA |
| 'Serialize Int32' | 646.4 ns | 44.55 ns | 26.51 ns | 0.16 | 0.01 | - | NA |
| 'Serialize 100 Messages (key+value)' | 32,499.2 ns | 304.97 ns | 181.49 ns | 7.86 | 0.17 | - | NA |
| 'ArrayBufferWriter + Copy' | 4,136.3 ns | 137.44 ns | 90.91 ns | 1.00 | 0.03 | - | NA |
| 'PooledBufferWriter Direct' | 3,529.3 ns | 312.95 ns | 186.23 ns | 0.85 | 0.05 | - | NA |
Compression Benchmarks
| Method | Mean | Error | StdDev | Allocated |
|---|---|---|---|---|
| 'Snappy Compress 1KB' | 13.908 μs | 1.872 μs | 1.1140 μs | - |
| 'Snappy Compress 1MB' | 926.587 μs | 11.170 μs | 5.8424 μs | - |
| 'Snappy Decompress 1KB' | 9.683 μs | 1.322 μs | 0.6917 μs | - |
| 'Snappy Decompress 1MB' | 2,059.509 μs | 748.319 μs | 494.9665 μs | - |
How to Read These Results
- Mean: Average execution time
- Error: Half of 99.9% confidence interval
- StdDev: Standard deviation of all measurements
- Ratio: Performance relative to baseline (Confluent.Kafka)
< 1.0= Dekaf is faster> 1.0= Confluent is faster1.0= Same performance
- Allocated: Heap memory allocated per operation
-= Zero allocations (ideal!)
Benchmarks are automatically run on every push to main.