Full benchmark run: i7-12800H, .NET 10

The first full run of the benchmark suite since the 0.1 release, with the --gate check. Analyzed on 2026-09-28.
CPU Intel Core i7-12800H (Alder Lake), pinned to logical core 0 (--affinity 1)
OS Windows 11 25H2 (10.0.26200)
Runtime .NET 10.0.11, x64 RyuJIT x86-64-v3 (SDK 10.0.400)
Tool BenchmarkDotNet 0.15.8, --filter '*' --memory --gate --affinity 1
Code 79ee062 on main, which is the 0.1.1 library code; the run started 4 minutes after that commit
Size 154 benchmarks in 40 minutes

Faster is how many times faster AvroSharp is than Apache.Avro: Apache.Avro's mean time divided by AvroSharp's. 2.00× means AvroSharp takes half the time.

Summary

  • The gate passes. Each of the 94 AvroSharp rows is faster than its Apache.Avro baseline, and none allocates more.
  • The multipliers range from 1.04× (bzip2 container writes) to 22.39× (xz container reads with generated code).
  • #32's "container benchmarks beat Apache for every codec" criterion is met on x64. Every codec is faster for both reads and writes, and in both the generic and the generated forms.
Area Faster
Varint decode 1.42–3.16×
Varint encode 1.09–4.52×
Binary encoding (longs, strings, mixed) 1.19–1.98×
Bulk long reads 2.02–3.27×
Schema parsing 1.08–1.58×
Records, generic model read 1.95×, write 3.93×
Records, generated code read 3.95×, write 6.18×
Schema resolution generic 1.32×, generated 2.72×
Showcase scenarios read 1.73–4.42×, write 3.77–20.20×
Container reads (null, deflate, snappy, zstandard) generic 2.90–3.72×, generated 5.98–7.36×
Container reads, xz generic 10.91–11.72×, generated 22.39×
Container reads, bzip2 1.63–2.11×
Container writes (null, deflate, snappy, zstandard, xz) generic 3.46–6.44×, generated 5.19–10.35×
Container writes, bzip2 1.04–1.05×

Where the margins are thin

  1. bzip2 container writes: 1.04–1.05×. Compression dominates the time: 125 ms out of 131 ms. Both libraries compress with SharpZipLib, so only the Avro encoding differs. More speed here would need a faster bzip2 compressor, not Avro changes.
  2. Varint encode of 3-byte values: 1.09×. Varint encodes of 4 and 5 bytes are 1.39× and 1.47×. Decode of 5-byte values is 1.42×. The mid-length varint paths remain the weakest spot (see hybrid varints).
  3. Small schema parsing: 1.08×. A small schema takes 7.3 µs against 7.9 µs. The large schema parses 1.58× faster, so the fixed cost of a parse is the part to look at. AvroSharp allocates about a third of Apache.Avro's memory either way.
  4. String decode and encode: 1.19× and 1.22×. The UTF-8 transcoding is the same work in both libraries. Decoding allocates the same memory, because the strings themselves are the allocation.

Other observations

  • AvroSharp allocates less everywhere. Writes allocate close to nothing: under 6 KB for a whole container file, against 5.7 MB for Apache.Avro, except xz at 777 KB. Generated reads allocate the least.
  • The generated code is about twice as fast as the generic model for records, resolution and container reads.
  • Pipelined async reads don't help on one core, which is expected. With --affinity 1, decompression and decoding cannot overlap. For fast codecs they are 3–10% slower than sequential reads. bzip2 is the exception: 2.04× against 1.65×. The pipelined numbers need a run without affinity to be meaningful.
  • Async and sync container reads and writes are within a few percent of each other.

Caveats

  • This is one machine: x64 on Windows with .NET 10. Arm64, Linux, and .NET 8 and 9 were not run.
  • Pinning to one core makes the numbers stable, but it hides any parallelism: the pipelined reads, and the thread pool behind async I/O.
  • The multipliers come from the means. The BenchmarkDotNet reports are in BenchmarkDotNet.Artifacts/results, which is not committed.

Results

Each table shows, for every case, the method (AvroSharp's side), Apache.Avro's mean, AvroSharp's mean, how many times faster AvroSharp is, and allocated memory per operation as AvroSharp / Apache.Avro.

Varints

Case Method Apache.Avro AvroSharp time Faster Allocated (AvroSharp / Apache)
Decode, Bytes 1 decode 122.82 µs 38.88 µs 3.16× 0 B / 88 B
Decode, Bytes 10 decode 623.31 µs 223.73 µs 2.79× 0 B / 89 B
Decode, Bytes 2 decode 173.77 µs 57.11 µs 3.04× 0 B / 88 B
Decode, Bytes 3 decode 235.63 µs 138.79 µs 1.70× 0 B / 89 B
Decode, Bytes 4 decode 304.71 µs 146.15 µs 2.08× 0 B / 89 B
Decode, Bytes 5 decode 343.54 µs 242.42 µs 1.42× 0 B / 89 B
Decode, Bytes 8 decode 513.58 µs 235.00 µs 2.19× 0 B / 89 B
Decode, Bytes Mixed1-10 decode 984.90 µs 523.16 µs 1.88× 0 B / 89 B
Decode, Bytes Mixed1-2 decode 434.16 µs 241.50 µs 1.80× 0 B / 88 B
Encode, Bytes 1 encode 63.27 µs 42.89 µs 1.48× 0 B / 0 B
Encode, Bytes 10 encode 697.92 µs 154.41 µs 4.52× 0 B / 1 B
Encode, Bytes 2 encode 119.54 µs 58.66 µs 2.04× 0 B / 0 B
Encode, Bytes 3 encode 174.24 µs 159.79 µs 1.09× 0 B / 0 B
Encode, Bytes 4 encode 234.55 µs 169.32 µs 1.39× 0 B / 0 B
Encode, Bytes 5 encode 278.02 µs 189.69 µs 1.47× 0 B / 0 B
Encode, Bytes 8 encode 553.04 µs 185.68 µs 2.98× 0 B / 1 B
Encode, Bytes Mixed1-10 encode 855.42 µs 411.19 µs 2.08× 0 B / 0 B
Encode, Bytes Mixed1-2 encode 344.04 µs 231.64 µs 1.49× 0 B / 0 B

Binary encoding

Case Method Apache.Avro AvroSharp time Faster Allocated (AvroSharp / Apache)
Decode, Workload Longs decode 983.1 µs 528.1 µs 1.86× 0 B / 88 B
Decode, Workload Mixed decode 4,581.4 µs 3,151.3 µs 1.45× 5032601 B / 7654133 B
Decode, Workload RealisticLongs decode 647.9 µs 387.9 µs 1.67× 0 B / 89 B
Decode, Workload Strings decode 2,718.4 µs 2,275.8 µs 1.19× 5032601 B / 5032688 B
Encode, Workload Longs encode 819.1 µs 413.9 µs 1.98× 0 B / 25 B
Encode, Workload Mixed encode 3,991.4 µs 2,626.1 µs 1.52× 1 B / 3612056 B
Encode, Workload RealisticLongs encode 571.4 µs 343.8 µs 1.66× 0 B / 24 B
Encode, Workload Strings encode 2,188.9 µs 1,800.2 µs 1.22× 1 B / 3612057 B

Bulk reads

Case Method Apache.Avro AvroSharp time Faster Allocated (AvroSharp / Apache)
Data Mixed read longs 244.99 µs 121.09 µs 2.02× 0 B / 88 B
Data Mixed scalar read long loop 244.99 µs 98.47 µs 2.49× 0 B / 88 B
Data SmallValues read longs 120.25 µs 36.73 µs 3.27× 0 B / 88 B
Data SmallValues scalar read long loop 120.25 µs 42.68 µs 2.82× 0 B / 88 B
Data Timestamps read longs 507.82 µs 236.19 µs 2.15× 0 B / 89 B
Data Timestamps scalar read long loop 507.82 µs 232.36 µs 2.19× 0 B / 89 B

Schema parsing

Case Method Apache.Avro AvroSharp time Faster Allocated (AvroSharp / Apache)
Parse, Schema Large parse 586.486 µs 371.223 µs 1.58× 439.21 KB / 1399.01 KB
Parse, Schema Small parse 7.861 µs 7.303 µs 1.08× 8.37 KB / 24.3 KB
ParseAndFingerprint, Schema Large parse and fingerprint 633.320 µs 406.204 µs 1.56× 511.07 KB / 1563.02 KB
ParseAndFingerprint, Schema Small parse and fingerprint 9.393 µs 7.024 µs 1.34× 9.85 KB / 27.81 KB

Records

Case Method Apache.Avro AvroSharp time Faster Allocated (AvroSharp / Apache)
Read read 1,455.2 ns 747.7 ns 1.95× 3424 B / 5032 B
Read generated read 1,455.2 ns 368.8 ns 3.95× 2192 B / 5032 B
Write write 1,290.3 ns 328.5 ns 3.93× 0 B / 5608 B
Write generated write 1,290.3 ns 208.7 ns 6.18× 0 B / 5608 B

Schema resolution

Case Method Apache.Avro AvroSharp time Faster Allocated (AvroSharp / Apache)
Read read 505.1 ns 383.7 ns 1.32× 784 B / 1632 B
Read generated read 505.1 ns 185.4 ns 2.72× 480 B / 1632 B

Showcase scenarios

Case Method Apache.Avro AvroSharp time Faster Allocated (AvroSharp / Apache)
Read, Scenario Counters read 229.73 ns 56.08 ns 4.10× 312 B / 688 B
Read, Scenario DoubleArray read 7,820.50 ns 2,777.05 ns 2.82× 16128 B / 32232 B
Read, Scenario IntArray read 5,612.59 ns 3,240.96 ns 1.73× 16128 B / 32232 B
Read, Scenario LongArray read 5,518.30 ns 3,066.80 ns 1.80× 16128 B / 32232 B
Read, Scenario Telemetry read 362.92 ns 82.14 ns 4.42× 456 B / 976 B
Write, Scenario Counters write 129.88 ns 34.44 ns 3.77× 0 B / 1048 B
Write, Scenario DoubleArray write 16,107.07 ns 797.19 ns 20.20× 0 B / 64024 B
Write, Scenario IntArray write 8,583.62 ns 1,056.97 ns 8.12× 0 B / 64024 B
Write, Scenario LongArray write 8,702.40 ns 1,044.06 ns 8.34× 0 B / 64024 B
Write, Scenario Telemetry write 396.77 ns 48.15 ns 8.24× 0 B / 1560 B

Container files

Case Method Apache.Avro AvroSharp time Faster Allocated (AvroSharp / Apache)
Read, Codec bzip2 read 33,170.0 µs 20,133.7 µs 1.65× 25637.82 KB / 31707.77 KB
Read, Codec bzip2 generated read 33,170.0 µs 15,686.6 µs 2.11× 24433.51 KB / 31707.77 KB
Read, Codec bzip2 read async 33,170.0 µs 20,299.2 µs 1.63× 25638.38 KB / 31707.77 KB
Read, Codec bzip2 read pipelined async 33,170.0 µs 16,223.6 µs 2.04× 25648.89 KB / 31707.77 KB
Read, Codec deflate read 2,615.2 µs 824.4 µs 3.17× 3357.23 KB / 8585.08 KB
Read, Codec deflate generated read 2,615.2 µs 437.1 µs 5.98× 2153.2 KB / 8585.08 KB
Read, Codec deflate read async 2,615.2 µs 857.6 µs 3.05× 3357.45 KB / 8585.08 KB
Read, Codec deflate read pipelined async 2,615.2 µs 900.7 µs 2.90× 3360.06 KB / 8585.08 KB
Read, Codec null read 2,586.4 µs 814.7 µs 3.17× 3355.79 KB / 8021.14 KB
Read, Codec null generated read 2,586.4 µs 402.4 µs 6.43× 2151.88 KB / 8021.14 KB
Read, Codec null read async 2,586.4 µs 817.9 µs 3.16× 3356.02 KB / 8021.14 KB
Read, Codec null read pipelined async 2,586.4 µs 859.7 µs 3.01× 3358.37 KB / 8021.14 KB
Read, Codec snappy read 3,142.8 µs 844.2 µs 3.72× 3356.26 KB / 8273 KB
Read, Codec snappy generated read 3,142.8 µs 426.8 µs 7.36× 2152.22 KB / 8273 KB
Read, Codec snappy read async 3,142.8 µs 873.1 µs 3.60× 3356.48 KB / 8273 KB
Read, Codec snappy read pipelined async 3,142.8 µs 889.0 µs 3.54× 3359.08 KB / 8273 KB
Read, Codec xz read 10,735.4 µs 918.4 µs 11.69× 3440.47 KB / 13706.05 KB
Read, Codec xz generated read 10,735.4 µs 479.4 µs 22.39× 2236.43 KB / 13706.05 KB
Read, Codec xz read async 10,735.4 µs 915.9 µs 11.72× 3440.69 KB / 13706.05 KB
Read, Codec xz read pipelined async 10,735.4 µs 984.4 µs 10.91× 3443.29 KB / 13706.05 KB
Read, Codec zstandard read 2,681.1 µs 809.0 µs 3.31× 3355.87 KB / 8584.18 KB
Read, Codec zstandard generated read 2,681.1 µs 412.9 µs 6.49× 2151.84 KB / 8584.18 KB
Read, Codec zstandard read async 2,681.1 µs 837.0 µs 3.20× 3356.09 KB / 8584.18 KB
Read, Codec zstandard read pipelined async 2,681.1 µs 855.2 µs 3.14× 3358.7 KB / 8584.18 KB
Write, Codec bzip2 write 131,395.2 µs 124,848.9 µs 1.05× 66915.67 KB / 72651.44 KB
Write, Codec bzip2 generated write 131,395.2 µs 126,538.4 µs 1.04× 66913.65 KB / 72651.44 KB
Write, Codec bzip2 write async 131,395.2 µs 125,161.6 µs 1.05× 66912.67 KB / 72651.44 KB
Write, Codec deflate write 1,687.4 µs 487.3 µs 3.46× 5.9 KB / 5721.89 KB
Write, Codec deflate generated write 1,687.4 µs 325.4 µs 5.19× 5.77 KB / 5721.89 KB
Write, Codec deflate write async 1,687.4 µs 488.1 µs 3.46× 5.9 KB / 5721.89 KB
Write, Codec null write 1,597.3 µs 372.7 µs 4.29× 4.7 KB / 5784.23 KB
Write, Codec null generated write 1,597.3 µs 224.0 µs 7.13× 4.6 KB / 5784.23 KB
Write, Codec null write async 1,597.3 µs 387.8 µs 4.12× 4.8 KB / 5784.23 KB
Write, Codec snappy write 2,048.5 µs 405.8 µs 5.05× 4.97 KB / 6347.28 KB
Write, Codec snappy generated write 2,048.5 µs 245.1 µs 8.36× 4.9 KB / 6347.28 KB
Write, Codec snappy write async 2,048.5 µs 411.7 µs 4.98× 5.03 KB / 6347.28 KB
Write, Codec xz write 11,675.7 µs 1,973.3 µs 5.92× 777.24 KB / 10849.34 KB
Write, Codec xz generated write 11,675.7 µs 1,746.6 µs 6.68× 777.31 KB / 10849.34 KB
Write, Codec xz write async 11,675.7 µs 2,009.4 µs 5.81× 777.31 KB / 10849.34 KB
Write, Codec zstandard write 2,828.8 µs 452.4 µs 6.25× 4.73 KB / 5721.35 KB
Write, Codec zstandard generated write 2,828.8 µs 273.2 µs 10.35× 4.67 KB / 5721.35 KB
Write, Codec zstandard write async 2,828.8 µs 439.0 µs 6.44× 4.8 KB / 5721.35 KB
Edit this page