Use ByteVectorChunk?

ByteVectorChunk is a type of Chunk[Byte] which is better optimized for avoiding copies. Based on our experience with byte copying in scalaz-stream, I have a feeling this will be a big win for us.

But it only exists in fs2-1.0-SNAPSHOT. We should think about how we want to cut over, or whether we want to kindly ask for an fs2-0.10 so we can start building on that.

[rossabaker]

Here’s a crude benchmark for concatenating some chunks and rendering it to an array for writing.

package org.http4s.bench

import fs2._
import fs2.interop.scodec.ByteVectorChunk
import org.openjdk.jmh.annotations._
import scodec.bits.ByteVector

@Fork(2)
@Threads(1)
@State(Scope.Benchmark)
@BenchmarkMode(Array(Mode.Throughput))
class BytesBench {
  val bs: Array[Byte] = "The quick brown fox jumped over the lazy dog.".getBytes("UTF-8")

  @Benchmark def byteVectorChunk = {
    val s = Segment.chunk(ByteVectorChunk(ByteVector.view(bs))) ++
      Segment.chunk(ByteVectorChunk(ByteVector.view(bs))) ++
      Segment.chunk(ByteVectorChunk(ByteVector.view(bs))) ++
      Segment.chunk(ByteVectorChunk(ByteVector.view(bs))) ++
      Segment.chunk(ByteVectorChunk(ByteVector.view(bs)))
    s.force.toArray
  }

  @Benchmark def bytes = {
    val s = Chunk.bytes(bs).toSegment ++
      Chunk.bytes(bs).toSegment ++
      Chunk.bytes(bs).toSegment ++
      Chunk.bytes(bs).toSegment ++
      Chunk.bytes(bs).toSegment
    s.force.toArray
  }
}

[info] Benchmark                    Mode  Cnt       Score       Error  Units
[info] BytesBench.byteVectorChunk  thrpt   40  350288.096 ± 12782.808  ops/s
[info] BytesBench.bytes            thrpt   40  470374.904 ± 19307.606  ops/s

[rossabaker]

Yes, it will.