[feature] Move to rayon backend (#7)

.github/workflows/ci.yaml

@@ -80,4 +80,4 @@ jobs:
         uses: Swatinem/rust-cache@v1
 
       - name: Run tests
-        run: cargo test --verbose --features parsnap_default,pargz_default
+        run: cargo test --verbose --features parsnap_default,pargz_default -- test_simple test_regression

Cargo.toml

@@ -15,6 +15,7 @@ description = "Parallel Gzip Compression"
 [lib]
 name = "gzp"
 path = "src/lib.rs"
+bench = false
 
 [features]
 default = []
@@ -31,11 +32,11 @@ rust_backend = ["flate2/miniz_oxide"]
 [dependencies]
 bytes = "1.0.1"
 flate2 = { version = "~1", default-features = false, optional = true }
-futures = "0.3.16"
 num_cpus = "1.13.0"
 thiserror = "1.0.26"
-tokio = { version = "1.9.0", features = ["sync", "rt-multi-thread"] }
 snap = { version = "~1", optional = true }
+flume = { version = "0.10.8", features = ["async"] }
+rayon = "1.5.1"
 
 [dev-dependencies]
 proptest = "1.0.0"

README.md

@@ -147,25 +147,10 @@ fn main() {
 
 ## Benchmarks
 
-All benchmarks were run on the file in `./bench-data/shakespeare.txt` catted together 100 times which creats a rough
+All benchmarks were run on the file in `./bench-data/shakespeare.txt` catted together 100 times which creates a rough
 550Mb file.
 
-Note that there are far more comprehensive comparisons of the tradeoffs in differet compression algorithms / compression
-levels elsewhere on the interent. This is meant to give rough understanding of the tradoffs involved. 
-
-| Name | Num Threads | Compression Level | Buffer Size | Time | File Size |
-| --- | - | ----------------- | -----------| ---- | --------- |
-| Gzip Only | NA | 3 | 128 Kb | 6.6s | 218 Mb |
-| Gzip | 1 | 3 | 128 Kb | 2.4s | 223 Mb |
-| Gzip | 4 | 3 | 128 Kb | 1.2s | 223 Mb |
-| Gzip | 8 | 3 | 128 Kb | 0.8s | 223 Mb |
-| Gzip | 16 | 3 | 128 Kb | 0.6s | 223 Mb |
-| Gzip | 30 | 3 | 128 Kb | 0.6s | 223 Mb |
-| Snap Only | NA | NA | 128 Kb | 1.6s | 333 Mb |
-| Snap | 1 | NA | 128 Kb | 0.7s | 333 Mb |
-| Snap | 4 | NA | 128 Kb | 0.5s | 333 Mb |
-| Snap | 8 | NA | 128 Kb | 0.4s | 333 Mb |
-| Snap | 16 | NA | 128 Kb | 0.4s | 333 Mb |
-| Snap | 30 | NA | 128 Kb | 0.4s | 333 Mb |
+The primary takeaway here is that you probably want to give `gzp` at least 4 threads. 2 threads breaks even with the
+overhead of orchestrating the multi-threadedness, 4 gives a roughly 3x improvement.
 
 ![benchmarks](./violin.svg)

benches/bench.rs

@@ -99,7 +99,7 @@ fn criterion_benchmark(c: &mut Criterion) {
     let buffersize = 64 * (1 << 10);
     let compression_level = 3;
     let mut group = c.benchmark_group("Compression");
-    for num_cpus in [1usize, 4, 8, 16, 30] {
+    for num_cpus in [2, 4, 8, 16, 30] {
         group.bench_with_input(
             BenchmarkId::new("Gzip", num_cpus),
             &num_cpus,