[3.3.0] Performance regression #2634
Description
This issue was originally created in fs2, so version numbers below refer to fs2.
I've observed a performance degradation in some scenarios on 3.2.3 version.
The throughput on small byte streams remained the same.
The throughput on bigger byte streams is decreased roughly by ~20%.
The memory allocation decreased by 10-15% on bigger byte streams.
See the benchmarks below for more details.
Stream usage
The project utilizes a TCP socket from the fs2-io module. I cannot share many details due to NDA, but the generalized usage of the socket is the following one:
val streamDecoder: StreamDecoder[Structure] =
StreamDecoder.many(StructureDecoder)
socket.reads
.through(streamDecoder.toPipeByte)
.evalMap(structure => queue.offer(Right(structure)))
.compile
.backgroundConsumed bytes per single invocation:
- createOne: 167 bytes
- returnRandomUUID: 135 bytes
- return100Record: 2683 bytes
3.2.2
Operation average time:
Benchmark Mode Cnt Score Error Units
DriverBenchmark.createOne avgt 25 4.725 ± 0.396 ms/op
DriverBenchmark.return100Records avgt 25 8.613 ± 0.488 ms/op
DriverBenchmark.returnRandomUUID avgt 25 3.542 ± 1.156 ms/op
Memory allocation:
DriverBenchmark.return100Records:·gc.alloc.rate 201.323 ± 11.576 MB/sec
DriverBenchmark.return100Records:·gc.alloc.rate.norm 3117144.635 ± 13036.746 B/op
DriverBenchmark.return100Records:·gc.churn.G1_Eden_Space 199.083 ± 23.093 MB/sec
DriverBenchmark.return100Records:·gc.churn.G1_Eden_Space.norm 3076695.652 ± 283692.848 B/op
DriverBenchmark.return100Records:·gc.churn.G1_Old_Gen 0.074 ± 0.113 MB/sec
DriverBenchmark.return100Records:·gc.churn.G1_Old_Gen.norm 1266.657 ± 1981.461 B/op
DriverBenchmark.return100Records:·gc.churn.G1_Survivor_Space 0.484 ± 0.444 MB/sec
DriverBenchmark.return100Records:·gc.churn.G1_Survivor_Space.norm 7614.493 ± 6998.312 B/op
DriverBenchmark.return100Records:·gc.count 72.000 counts
DriverBenchmark.return100Records:·gc.time 300.000 ms
3.2.3
Operation average time:
Benchmark Mode Cnt Score Error Units
DriverBenchmark.createOne avgt 25 4.862 ± 0.414 ms/op
DriverBenchmark.return100Records avgt 25 11.008 ± 0.356 ms/op
DriverBenchmark.returnRandomUUID avgt 25 3.068 ± 0.299 ms/op
Memory allocation:
DriverBenchmark.return100Records:·gc.alloc.rate 169.961 ± 6.376 MB/sec
DriverBenchmark.return100Records:·gc.alloc.rate.norm 3383113.516 ± 12432.017 B/op
DriverBenchmark.return100Records:·gc.churn.G1_Eden_Space 175.288 ± 24.073 MB/sec
DriverBenchmark.return100Records:·gc.churn.G1_Eden_Space.norm 3484911.125 ± 432781.648 B/op
DriverBenchmark.return100Records:·gc.churn.G1_Old_Gen 0.061 ± 0.075 MB/sec
DriverBenchmark.return100Records:·gc.churn.G1_Old_Gen.norm 1224.015 ± 1529.533 B/op
DriverBenchmark.return100Records:·gc.churn.G1_Survivor_Space 0.484 ± 0.566 MB/sec
DriverBenchmark.return100Records:·gc.churn.G1_Survivor_Space.norm 9489.291 ± 11071.370 B/op
DriverBenchmark.return100Records:·gc.count 56.000 counts
DriverBenchmark.return100Records:·gc.time 433.000 ms