[
https://issues.apache.org/jira/browse/CASSANDRA-15215?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17440071#comment-17440071
]
Aleksandr Sorokoumov commented on CASSANDRA-15215:
--------------------------------------------------
Short status update: I just pushed changes to the benchmarks as suggested by
Branimir and Benedict. The changes are:
* All tests have monomorphic, bimorphic, and megamorphic versions.
* Added a test that writes random longs to the ByteBuffer.
* Added a test for calculating VInt size and applied Branimir's formula.
Results:
h4. Baseline
{noformat}
Benchmark (allocation) Mode Cnt Score
Error Units
VIntCodingBench.testComputeUnsignedVIntSize monomorphic avgt 15 17.069 ±
1.087 ns/op
VIntCodingBench.testComputeUnsignedVIntSize bimorphic avgt 15 17.323 ±
0.656 ns/op
VIntCodingBench.testComputeUnsignedVIntSize megamorphic avgt 15 16.791 ±
0.473 ns/op
VIntCodingBench.testWrite1Byte monomorphic avgt 15 9.047 ±
0.254 ns/op
VIntCodingBench.testWrite1Byte bimorphic avgt 15 16.935 ±
0.207 ns/op
VIntCodingBench.testWrite1Byte megamorphic avgt 15 17.835 ±
0.090 ns/op
VIntCodingBench.testWrite2Bytes monomorphic avgt 15 18.612 ±
0.194 ns/op
VIntCodingBench.testWrite2Bytes bimorphic avgt 15 25.033 ±
0.239 ns/op
VIntCodingBench.testWrite2Bytes megamorphic avgt 15 28.352 ±
0.115 ns/op
VIntCodingBench.testWrite3Bytes monomorphic avgt 15 21.333 ±
0.197 ns/op
VIntCodingBench.testWrite3Bytes bimorphic avgt 15 26.173 ±
0.170 ns/op
VIntCodingBench.testWrite3Bytes megamorphic avgt 15 29.983 ±
0.208 ns/op
VIntCodingBench.testWrite4Bytes monomorphic avgt 15 21.229 ±
0.245 ns/op
VIntCodingBench.testWrite4Bytes bimorphic avgt 15 28.966 ±
0.606 ns/op
VIntCodingBench.testWrite4Bytes megamorphic avgt 15 33.219 ±
1.276 ns/op
VIntCodingBench.testWrite5Bytes monomorphic avgt 15 22.886 ±
0.602 ns/op
VIntCodingBench.testWrite5Bytes bimorphic avgt 15 29.209 ±
1.077 ns/op
VIntCodingBench.testWrite5Bytes megamorphic avgt 15 32.731 ±
0.944 ns/op
VIntCodingBench.testWrite6Bytes monomorphic avgt 15 22.579 ±
0.794 ns/op
VIntCodingBench.testWrite6Bytes bimorphic avgt 15 29.067 ±
0.678 ns/op
VIntCodingBench.testWrite6Bytes megamorphic avgt 15 35.419 ±
1.496 ns/op
VIntCodingBench.testWrite7Bytes monomorphic avgt 15 22.823 ±
0.527 ns/op
VIntCodingBench.testWrite7Bytes bimorphic avgt 15 29.521 ±
1.216 ns/op
VIntCodingBench.testWrite7Bytes megamorphic avgt 15 34.295 ±
2.327 ns/op
VIntCodingBench.testWrite8Bytes monomorphic avgt 15 22.032 ±
0.918 ns/op
VIntCodingBench.testWrite8Bytes bimorphic avgt 15 30.388 ±
1.015 ns/op
VIntCodingBench.testWrite8Bytes megamorphic avgt 15 33.632 ±
1.200 ns/op
VIntCodingBench.testWrite9Bytes monomorphic avgt 15 22.616 ±
1.309 ns/op
VIntCodingBench.testWrite9Bytes bimorphic avgt 15 29.291 ±
1.096 ns/op
VIntCodingBench.testWrite9Bytes megamorphic avgt 15 32.597 ±
0.807 ns/op
VIntCodingBench.testWriteRandomLong monomorphic avgt 15 35.010 ±
1.145 ns/op
VIntCodingBench.testWriteRandomLong bimorphic avgt 15 43.090 ±
0.615 ns/op
VIntCodingBench.testWriteRandomLong megamorphic avgt 15 43.196 ±
1.742 ns/op
{noformat}
h4. Patch
{noformat}
VIntCodingBench.testComputeUnsignedVIntSize monomorphic avgt 15 16.339 ±
0.418 ns/op
VIntCodingBench.testComputeUnsignedVIntSize bimorphic avgt 15 16.340 ±
0.417 ns/op
VIntCodingBench.testComputeUnsignedVIntSize megamorphic avgt 15 16.435 ±
0.408 ns/op
VIntCodingBench.testWrite1Byte monomorphic avgt 15 9.362 ±
0.208 ns/op
VIntCodingBench.testWrite1Byte bimorphic avgt 15 18.164 ±
0.839 ns/op
VIntCodingBench.testWrite1Byte megamorphic avgt 15 19.800 ±
0.942 ns/op
VIntCodingBench.testWrite2Bytes monomorphic avgt 15 10.094 ±
0.444 ns/op
VIntCodingBench.testWrite2Bytes bimorphic avgt 15 18.310 ±
0.813 ns/op
VIntCodingBench.testWrite2Bytes megamorphic avgt 15 19.685 ±
0.692 ns/op
VIntCodingBench.testWrite3Bytes monomorphic avgt 15 11.541 ±
0.433 ns/op
VIntCodingBench.testWrite3Bytes bimorphic avgt 15 19.087 ±
0.720 ns/op
VIntCodingBench.testWrite3Bytes megamorphic avgt 15 20.518 ±
1.035 ns/op
VIntCodingBench.testWrite4Bytes monomorphic avgt 15 10.677 ±
0.417 ns/op
VIntCodingBench.testWrite4Bytes bimorphic avgt 15 18.815 ±
0.921 ns/op
VIntCodingBench.testWrite4Bytes megamorphic avgt 15 21.106 ±
0.731 ns/op
VIntCodingBench.testWrite5Bytes monomorphic avgt 15 11.564 ±
0.549 ns/op
VIntCodingBench.testWrite5Bytes bimorphic avgt 15 18.961 ±
0.783 ns/op
VIntCodingBench.testWrite5Bytes megamorphic avgt 15 20.939 ±
0.541 ns/op
VIntCodingBench.testWrite6Bytes monomorphic avgt 15 13.463 ±
0.441 ns/op
VIntCodingBench.testWrite6Bytes bimorphic avgt 15 20.188 ±
0.792 ns/op
VIntCodingBench.testWrite6Bytes megamorphic avgt 15 23.087 ±
1.044 ns/op
VIntCodingBench.testWrite7Bytes monomorphic avgt 15 12.655 ±
0.347 ns/op
VIntCodingBench.testWrite7Bytes bimorphic avgt 15 21.307 ±
0.892 ns/op
VIntCodingBench.testWrite7Bytes megamorphic avgt 15 23.209 ±
0.526 ns/op
VIntCodingBench.testWrite8Bytes monomorphic avgt 15 11.712 ±
0.404 ns/op
VIntCodingBench.testWrite8Bytes bimorphic avgt 15 19.179 ±
0.730 ns/op
VIntCodingBench.testWrite8Bytes megamorphic avgt 15 20.692 ±
0.627 ns/op
VIntCodingBench.testWrite9Bytes monomorphic avgt 15 12.029 ±
0.387 ns/op
VIntCodingBench.testWrite9Bytes bimorphic avgt 15 19.826 ±
0.971 ns/op
VIntCodingBench.testWrite9Bytes megamorphic avgt 15 21.953 ±
0.450 ns/op
VIntCodingBench.testWriteRandomLong monomorphic avgt 15 24.668 ±
0.523 ns/op
VIntCodingBench.testWriteRandomLong bimorphic avgt 15 31.157 ±
0.832 ns/op
VIntCodingBench.testWriteRandomLong megamorphic avgt 15 32.698 ±
1.376 ns/op
{noformat}
Next week I plan to start working on {{{}DataOutputPlus#writeBytes(long
register, int bytes){}}}.
> VIntCoding should read and write more efficiently
> -------------------------------------------------
>
> Key: CASSANDRA-15215
> URL: https://issues.apache.org/jira/browse/CASSANDRA-15215
> Project: Cassandra
> Issue Type: Bug
> Components: Local/Compaction, Local/SSTable
> Reporter: Benedict Elliott Smith
> Assignee: Aleksandr Sorokoumov
> Priority: Normal
> Fix For: 3.0.x, 3.11.x, 4.x
>
>
> Most vints occupy significantly fewer than 8 bytes, and most buffers have >=
> 8 bytes spare, in which case we can construct the relevant bytes in a
> register and memcpy them to the correct position. Since we read and write a
> lot of vints, this waste is probably measurable, particularly during
> compaction and flush, and can probably be considered a performance bug.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
