st3 0.4.1

A very fast lock-free, bounded, work-stealing LIFO queue.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242

use std::iter;
use std::sync::{Arc, Condvar, Mutex};
use std::thread::spawn;
use std::time::Instant;

use criterion::{criterion_group, criterion_main, Criterion};
use num_cpus;
use st3;

mod generic_queue;
mod tokio_queue;

use generic_queue::{
    CrossbeamFifoWorker, CrossbeamLifoWorker, GenericStealError, GenericStealer, GenericWorker,
};

// Single-threaded benchmark.
//
// `N` items are pushed and then popped from the queue.
pub fn push_pop<T, W: GenericWorker<usize>, const N: usize>(name: &str, c: &mut Criterion) {
    let worker = W::new();
    c.bench_function(&format!("push_pop-{}", name), |b| {
        b.iter(|| {
            for i in 0..N {
                let _ = worker.push(i);
            }
            for _ in 0..N {
                let _ = worker.pop();
            }
        })
    });
}

pub fn push_pop_small_batch_st3_fifo(c: &mut Criterion) {
    push_pop::<usize, st3::fifo::Worker<_>, 64>("small_batch-st3_fifo", c);
}
pub fn push_pop_small_batch_st3_lifo(c: &mut Criterion) {
    push_pop::<usize, st3::lifo::Worker<_>, 64>("small_batch-st3_lifo", c);
}
pub fn push_pop_small_batch_tokio(c: &mut Criterion) {
    push_pop::<usize, tokio_queue::Local<_>, 64>("small_batch-tokio", c);
}
pub fn push_pop_small_batch_crossbeam_fifo(c: &mut Criterion) {
    push_pop::<usize, CrossbeamFifoWorker<_>, 64>("small_batch-crossbeam_fifo", c);
}
pub fn push_pop_small_batch_crossbeam_lifo(c: &mut Criterion) {
    push_pop::<usize, CrossbeamLifoWorker<_>, 64>("small_batch-crossbeam_lifo", c);
}
pub fn push_pop_large_batch_st3_fifo(c: &mut Criterion) {
    push_pop::<usize, st3::fifo::Worker<_>, 256>("large_batch-st3_fifo", c);
}
pub fn push_pop_large_batch_st3_lifo(c: &mut Criterion) {
    push_pop::<usize, st3::lifo::Worker<_>, 256>("large_batch-st3_lifo", c);
}
pub fn push_pop_large_batch_tokio(c: &mut Criterion) {
    push_pop::<usize, tokio_queue::Local<_>, 256>("large_batch-tokio", c);
}
pub fn push_pop_large_batch_crossbeam_fifo(c: &mut Criterion) {
    push_pop::<usize, CrossbeamFifoWorker<_>, 256>("large_batch-crossbeam_fifo", c);
}
pub fn push_pop_large_batch_crossbeam_lifo(c: &mut Criterion) {
    push_pop::<usize, CrossbeamLifoWorker<_>, 256>("large_batch-crossbeam_lifo", c);
}

// Multi-threaded synthetic benchmark.
//
// This is a toy work-stealing executor which schedules a random number of tasks
// (up to 256), each one being repeated by re-injection onto its worker a random
// number of times (up to 100). The number of tasks initially assigned to each
// workers and the number of times each task is to be repeated are
// deterministically pre-determined, meaning that the workload is the same for
// all benchmarked queues. All queues use the crossbeam work-stealing strategy:
// half of the tasks are stolen, up to a maximum of 32 tasks. Nevertheless, the
// re-distribution of tasks via work-stealing is ultimately non-deterministic as
// it is affected by thread timing. The numbers from this benchmark must
// therefore be taken with a grain of salt.
pub fn executor<T, W: GenericWorker<u64> + 'static>(name: &str, c: &mut Criterion) {
    c.bench_function(&format!("executor-{}", name), |b| {
        // The maximum number of tasks to be enqueued.
        const MAX_TASKS_PER_THREAD: u64 = 256;
        // The maximum number of times a task may be re-injected onto a worker.
        const MAX_TASK_REPEAT: u64 = 100;

        // Use up to 4 threads, depending on the number of CPU threads.
        let thread_count = num_cpus::get().min(4);

        b.iter(|| {
            let workers: Vec<_> = iter::repeat_with(|| W::new()).take(thread_count).collect();
            let stealers: Vec<_> = workers.iter().map(|w| w.stealer()).collect();
            // A barrier that block all threads until all are ready.
            let wait_barrier = Arc::new((Mutex::new(0usize), Condvar::new()));
            let mut threads = Vec::with_capacity(thread_count);

            for (th_id, worker) in workers.into_iter().enumerate() {
                let stealers = stealers.clone();
                let wait_barrier = wait_barrier.clone();

                threads.push(spawn(move || -> Option<Instant> {
                    // Seed the RNG with the thread ID.
                    let mut rng = oorandom::Rand64::new(th_id as u128);

                    // Schedule from 1 to MAX_TASKS_PER_THREAD tasks.
                    let task_count = rng.rand_range(1..(MAX_TASKS_PER_THREAD + 1));
                    for _ in 0..task_count {
                        worker
                            .push(rng.rand_range(0..MAX_TASK_REPEAT))
                            .expect("attempting to schedule more tasks than the queue can hold");
                    }

                    // Preallocate an array that will be populated with the IDs
                    // of other workers.
                    let mut other_workers_id = Vec::with_capacity(thread_count - 1);
                    other_workers_id.resize(thread_count - 1, 0);

                    // Block until all threads are ready.
                    let start_time = {
                        let (lock, cvar) = &*wait_barrier;
                        let mut ready_count = lock.lock().unwrap();
                        *ready_count += 1;
                        if *ready_count == thread_count {
                            cvar.notify_all();

                            Some(Instant::now())
                        } else {
                            // Block until all threads are ready.
                            while *ready_count < thread_count {
                                ready_count = cvar.wait(ready_count).unwrap();
                            }

                            None
                        }
                    };

                    // Loop over all tasks in the queue and enqueue them again
                    // after decreasing their repeat count, until the count
                    // falls to 0.
                    'new_task: loop {
                        if let Some(repeat_count) = worker.pop() {
                            // Re-inject the task if needed.
                            if repeat_count > 0 {
                                // The actual queue size should never be greater
                                // than `MAX_TASKS_PER_THREAD`. Unfortunately
                                // the tokio queue can spuriously report a full
                                // queue on a push when many pop/push operations
                                // are made while a stealer is preempted. In
                                // such rare cases, spinning is necessary.
                                while let Err(_) = worker.push(repeat_count - 1) {}
                            }
                        } else {
                            // No more local tasks, try to steal.
                            // Make an array containing all other thread IDs.
                            for (i, th) in other_workers_id.iter_mut().enumerate() {
                                if i < th_id {
                                    *th = i;
                                } else {
                                    *th = i + 1;
                                }
                            }

                            // Randomly select a worker to steal from. Upon
                            // failure, try again until all workers have been
                            // found empty.
                            let mut pool_size = thread_count - 1;
                            loop {
                                // Choose randomly another worker to steal from.
                                let idx = rng.rand_range(0..pool_size as u64) as usize;
                                let steal_from = other_workers_id[idx];
                                match stealers[steal_from].steal_batch_and_pop(&worker) {
                                    Ok(repeat_count) => {
                                        // Re-inject the task if needed.
                                        if repeat_count > 0 {
                                            while let Err(_) = worker.push(repeat_count - 1) {}
                                        }
                                        continue 'new_task;
                                    }
                                    Err(GenericStealError::Empty) => {
                                        // Give up if all workers were found
                                        // empty when attempting to steal and
                                        // this was the last worker.
                                        if pool_size == 1 {
                                            return start_time;
                                        }
                                        // Evict the sampled worker from the
                                        // pool and replace it with the last
                                        // worker in the list, then try again
                                        // to steal within this reduced pool.
                                        pool_size -= 1;
                                        other_workers_id[idx] = other_workers_id[pool_size];
                                    }
                                    Err(GenericStealError::Busy) => {}
                                }
                            }
                        }
                    }
                }));
            }
            let mut start = None;
            for th in threads {
                if let Ok(Some(start_time)) = th.join() {
                    start = Some(start_time);
                }
            }

            start.unwrap().elapsed()
        })
    });
}
pub fn executor_st3_fifo(c: &mut Criterion) {
    executor::<usize, st3::fifo::Worker<_>>("st3_fifo", c);
}
pub fn executor_st3_lifo(c: &mut Criterion) {
    executor::<usize, st3::lifo::Worker<_>>("st3_lifo", c);
}
pub fn executor_tokio(c: &mut Criterion) {
    executor::<usize, tokio_queue::Local<_>>("tokio", c);
}
pub fn executor_crossbeam_fifo(c: &mut Criterion) {
    executor::<usize, CrossbeamFifoWorker<_>>("crossbeam_fifo", c);
}
pub fn executor_crossbeam_lifo(c: &mut Criterion) {
    executor::<usize, CrossbeamLifoWorker<_>>("crossbeam_lifo", c);
}

criterion_group!(
    benches,
    push_pop_small_batch_st3_fifo,
    push_pop_small_batch_st3_lifo,
    push_pop_small_batch_tokio,
    push_pop_small_batch_crossbeam_fifo,
    push_pop_small_batch_crossbeam_lifo,
    push_pop_large_batch_st3_fifo,
    push_pop_large_batch_st3_lifo,
    push_pop_large_batch_tokio,
    push_pop_large_batch_crossbeam_fifo,
    push_pop_large_batch_crossbeam_lifo,
    executor_st3_fifo,
    executor_st3_lifo,
    executor_tokio,
    executor_crossbeam_fifo,
    executor_crossbeam_lifo,
);
criterion_main!(benches);