Files
rustfs/ecstore/benches/comparison_benchmark.rs
T
2025-06-10 00:09:05 +08:00

331 lines
12 KiB
Rust

//! 专门比较 Pure Erasure 和 Hybrid (SIMD) 模式性能的基准测试
//!
//! 这个基准测试使用不同的feature编译配置来直接对比两种实现的性能。
//!
//! ## 运行比较测试
//!
//! ```bash
//! # 测试 Pure Erasure 实现 (默认)
//! cargo bench --bench comparison_benchmark
//!
//! # 测试 Hybrid (SIMD) 实现
//! cargo bench --bench comparison_benchmark --features reed-solomon-simd
//!
//! # 测试强制 erasure-only 模式
//! cargo bench --bench comparison_benchmark --features reed-solomon-erasure
//!
//! # 生成对比报告
//! cargo bench --bench comparison_benchmark -- --save-baseline erasure
//! cargo bench --bench comparison_benchmark --features reed-solomon-simd -- --save-baseline hybrid
//! ```
use criterion::{BenchmarkId, Criterion, Throughput, black_box, criterion_group, criterion_main};
use ecstore::erasure_coding::Erasure;
use std::time::Duration;
/// 基准测试数据配置
struct TestData {
data: Vec<u8>,
size_name: &'static str,
}
impl TestData {
fn new(size: usize, size_name: &'static str) -> Self {
let data = (0..size).map(|i| (i % 256) as u8).collect();
Self { data, size_name }
}
}
/// 生成不同大小的测试数据集
fn generate_test_datasets() -> Vec<TestData> {
vec![
TestData::new(1024, "1KB"), // 小数据
TestData::new(8 * 1024, "8KB"), // 中小数据
TestData::new(64 * 1024, "64KB"), // 中等数据
TestData::new(256 * 1024, "256KB"), // 中大数据
TestData::new(1024 * 1024, "1MB"), // 大数据
TestData::new(4 * 1024 * 1024, "4MB"), // 超大数据
]
}
/// 编码性能比较基准测试
fn bench_encode_comparison(c: &mut Criterion) {
let datasets = generate_test_datasets();
let configs = vec![
(4, 2, "4+2"), // 常用配置
(6, 3, "6+3"), // 50%冗余
(8, 4, "8+4"), // 50%冗余,更多分片
];
for dataset in &datasets {
for (data_shards, parity_shards, config_name) in &configs {
let test_name = format!("{}_{}_{}", dataset.size_name, config_name, get_implementation_name());
let mut group = c.benchmark_group("encode_comparison");
group.throughput(Throughput::Bytes(dataset.data.len() as u64));
group.sample_size(20);
group.measurement_time(Duration::from_secs(10));
// 检查是否能够创建erasure实例(某些配置在纯SIMD模式下可能失败)
match Erasure::new(*data_shards, *parity_shards, dataset.data.len()).encode_data(&dataset.data) {
Ok(_) => {
group.bench_with_input(
BenchmarkId::new("implementation", &test_name),
&(&dataset.data, *data_shards, *parity_shards),
|b, (data, data_shards, parity_shards)| {
let erasure = Erasure::new(*data_shards, *parity_shards, data.len());
b.iter(|| {
let shards = erasure.encode_data(black_box(data)).unwrap();
black_box(shards);
});
},
);
}
Err(e) => {
println!("⚠️ 跳过测试 {} - 配置不支持: {}", test_name, e);
}
}
group.finish();
}
}
}
/// 解码性能比较基准测试
fn bench_decode_comparison(c: &mut Criterion) {
let datasets = generate_test_datasets();
let configs = vec![(4, 2, "4+2"), (6, 3, "6+3"), (8, 4, "8+4")];
for dataset in &datasets {
for (data_shards, parity_shards, config_name) in &configs {
let test_name = format!("{}_{}_{}", dataset.size_name, config_name, get_implementation_name());
let erasure = Erasure::new(*data_shards, *parity_shards, dataset.data.len());
// 预先编码数据 - 检查是否支持此配置
match erasure.encode_data(&dataset.data) {
Ok(encoded_shards) => {
let mut group = c.benchmark_group("decode_comparison");
group.throughput(Throughput::Bytes(dataset.data.len() as u64));
group.sample_size(20);
group.measurement_time(Duration::from_secs(10));
group.bench_with_input(
BenchmarkId::new("implementation", &test_name),
&(&encoded_shards, *data_shards, *parity_shards),
|b, (shards, data_shards, parity_shards)| {
let erasure = Erasure::new(*data_shards, *parity_shards, dataset.data.len());
b.iter(|| {
// 模拟最大可恢复的数据丢失
let mut shards_opt: Vec<Option<Vec<u8>>> =
shards.iter().map(|shard| Some(shard.to_vec())).collect();
// 丢失等于奇偶校验分片数量的分片
for item in shards_opt.iter_mut().take(*parity_shards) {
*item = None;
}
erasure.decode_data(black_box(&mut shards_opt)).unwrap();
black_box(&shards_opt);
});
},
);
group.finish();
}
Err(e) => {
println!("⚠️ 跳过解码测试 {} - 配置不支持: {}", test_name, e);
}
}
}
}
}
/// 分片大小敏感性测试
fn bench_shard_size_sensitivity(c: &mut Criterion) {
let data_shards = 4;
let parity_shards = 2;
// 测试不同的分片大小,特别关注SIMD的临界点
let shard_sizes = vec![32, 64, 128, 256, 512, 1024, 2048, 4096, 8192];
let mut group = c.benchmark_group("shard_size_sensitivity");
group.sample_size(15);
group.measurement_time(Duration::from_secs(8));
for shard_size in shard_sizes {
let total_size = shard_size * data_shards;
let data = (0..total_size).map(|i| (i % 256) as u8).collect::<Vec<u8>>();
let test_name = format!("{}B_shard_{}", shard_size, get_implementation_name());
group.throughput(Throughput::Bytes(total_size as u64));
// 检查此分片大小是否支持
let erasure = Erasure::new(data_shards, parity_shards, data.len());
match erasure.encode_data(&data) {
Ok(_) => {
group.bench_with_input(BenchmarkId::new("shard_size", &test_name), &data, |b, data| {
let erasure = Erasure::new(data_shards, parity_shards, data.len());
b.iter(|| {
let shards = erasure.encode_data(black_box(data)).unwrap();
black_box(shards);
});
});
}
Err(e) => {
println!("⚠️ 跳过分片大小测试 {} - 不支持: {}", test_name, e);
}
}
}
group.finish();
}
/// 高负载并发测试
fn bench_concurrent_load(c: &mut Criterion) {
use std::sync::Arc;
use std::thread;
let data_size = 1024 * 1024; // 1MB
let data = Arc::new((0..data_size).map(|i| (i % 256) as u8).collect::<Vec<u8>>());
let erasure = Arc::new(Erasure::new(4, 2, data_size));
let mut group = c.benchmark_group("concurrent_load");
group.throughput(Throughput::Bytes(data_size as u64));
group.sample_size(10);
group.measurement_time(Duration::from_secs(15));
let test_name = format!("1MB_concurrent_{}", get_implementation_name());
group.bench_function(&test_name, |b| {
b.iter(|| {
let handles: Vec<_> = (0..4)
.map(|_| {
let data_clone = data.clone();
let erasure_clone = erasure.clone();
thread::spawn(move || {
let shards = erasure_clone.encode_data(&data_clone).unwrap();
black_box(shards);
})
})
.collect();
for handle in handles {
handle.join().unwrap();
}
});
});
group.finish();
}
/// 错误恢复能力测试
fn bench_error_recovery_performance(c: &mut Criterion) {
let data_size = 256 * 1024; // 256KB
let data = (0..data_size).map(|i| (i % 256) as u8).collect::<Vec<u8>>();
let configs = vec![
(4, 2, 1), // 丢失1个分片
(4, 2, 2), // 丢失2个分片(最大可恢复)
(6, 3, 2), // 丢失2个分片
(6, 3, 3), // 丢失3个分片(最大可恢复)
(8, 4, 3), // 丢失3个分片
(8, 4, 4), // 丢失4个分片(最大可恢复)
];
let mut group = c.benchmark_group("error_recovery");
group.throughput(Throughput::Bytes(data_size as u64));
group.sample_size(15);
group.measurement_time(Duration::from_secs(8));
for (data_shards, parity_shards, lost_shards) in configs {
let erasure = Erasure::new(data_shards, parity_shards, data_size);
let test_name = format!("{}+{}_lost{}_{}", data_shards, parity_shards, lost_shards, get_implementation_name());
// 检查此配置是否支持
match erasure.encode_data(&data) {
Ok(encoded_shards) => {
group.bench_with_input(
BenchmarkId::new("recovery", &test_name),
&(&encoded_shards, data_shards, parity_shards, lost_shards),
|b, (shards, data_shards, parity_shards, lost_shards)| {
let erasure = Erasure::new(*data_shards, *parity_shards, data_size);
b.iter(|| {
let mut shards_opt: Vec<Option<Vec<u8>>> = shards.iter().map(|shard| Some(shard.to_vec())).collect();
// 丢失指定数量的分片
for item in shards_opt.iter_mut().take(*lost_shards) {
*item = None;
}
erasure.decode_data(black_box(&mut shards_opt)).unwrap();
black_box(&shards_opt);
});
},
);
}
Err(e) => {
println!("⚠️ 跳过错误恢复测试 {} - 配置不支持: {}", test_name, e);
}
}
}
group.finish();
}
/// 内存效率测试
fn bench_memory_efficiency(c: &mut Criterion) {
let data_shards = 4;
let parity_shards = 2;
let data_size = 1024 * 1024; // 1MB
let mut group = c.benchmark_group("memory_efficiency");
group.throughput(Throughput::Bytes(data_size as u64));
group.sample_size(10);
group.measurement_time(Duration::from_secs(8));
let test_name = format!("memory_pattern_{}", get_implementation_name());
// 测试连续多次编码对内存的影响
group.bench_function(format!("{}_continuous", test_name), |b| {
let erasure = Erasure::new(data_shards, parity_shards, data_size);
b.iter(|| {
for i in 0..10 {
let data = vec![(i % 256) as u8; data_size];
let shards = erasure.encode_data(black_box(&data)).unwrap();
black_box(shards);
}
});
});
// 测试大量小编码任务
group.bench_function(format!("{}_small_chunks", test_name), |b| {
let chunk_size = 1024; // 1KB chunks
let erasure = Erasure::new(data_shards, parity_shards, chunk_size);
b.iter(|| {
for i in 0..1024 {
let data = vec![(i % 256) as u8; chunk_size];
let shards = erasure.encode_data(black_box(&data)).unwrap();
black_box(shards);
}
});
});
group.finish();
}
/// 获取当前实现的名称
fn get_implementation_name() -> &'static str {
#[cfg(feature = "reed-solomon-simd")]
return "hybrid";
#[cfg(not(feature = "reed-solomon-simd"))]
return "erasure";
}
criterion_group!(
benches,
bench_encode_comparison,
bench_decode_comparison,
bench_shard_size_sensitivity,
bench_concurrent_load,
bench_error_recovery_performance,
bench_memory_efficiency
);
criterion_main!(benches);