breve 0.1.0

In-memory cache implementation with Uno as the admission policy and S3-FIFO as the eviction policy
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

// Copyright 2025 Chojan Shang.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

use std::hash::Hash;
use std::sync::atomic::{AtomicUsize, Ordering};

use crate::atomic::AtomicF32;
use crate::estimator::Estimator;

// bare-minimum UnoSketch, based on Count-Min Sketch, to estimate freq and reuse distance in a limited window.
pub(crate) struct UnoSketch {
    freq_estimator: Estimator,
    reuse_estimator: Estimator,
    window_counter: AtomicUsize,
    cache_size: usize,
    window_limit: usize,
}

impl UnoSketch {
    pub fn get<T: Hash + Clone>(&self, key: T) -> (u8, usize) {
        let freq = self.freq_estimator.get(key.clone());
        let reuse_freq = self.reuse_estimator.get(key);
        let reuse = self.reuse_estimate(reuse_freq);
        (freq, reuse)
    }

    pub fn tick(&self) {
        let window_size = self.window_counter.fetch_add(1, Ordering::SeqCst);
        if window_size % (self.cache_size * 2) == 0 {
            self.freq_estimator.exponential_decay(0.8);
        }
        // When window_size concurrently increases, only one resets the window and age the estimator.
        // > self.window_limit * 2 is a safety net in case for whatever reason window_size grows
        // out of control
        if window_size == self.window_limit || window_size > self.window_limit * 2 {
            self.window_counter.store(0, Ordering::SeqCst);
            self.reuse_estimator.age(1); // right shift, next window
        }
    }

    pub fn incr<T: Hash + Clone>(&self, key: T) -> (u8, usize) {
        self.tick();
        let freq = self.freq_estimator.incr(key.clone());
        let reuse_freq = self.reuse_estimator.incr(key);
        let reuse = self.reuse_estimate(reuse_freq);
        (freq, reuse)
    }

    pub fn incr_freq_only<T: Hash + Clone>(&self, key: T) -> (u8, usize) {
        let freq = self.freq_estimator.incr(key.clone());
        let reuse_freq = self.reuse_estimator.get(key);
        let reuse = self.reuse_estimate(reuse_freq);
        (freq, reuse)
    }

    pub fn incr_reuse_only<T: Hash + Clone>(&self, key: T) -> (u8, usize) {
        let freq = self.freq_estimator.get(key.clone());
        let reuse_freq = self.reuse_estimator.incr(key);
        let reuse = self.reuse_estimate(reuse_freq);
        (freq, reuse)
    }

    // because we use 8-bits counters, window size can be 256 * the cache size
    pub fn new(cache_size: usize) -> Self {
        Self {
            freq_estimator: Estimator::optimal(cache_size),
            reuse_estimator: Estimator::optimal(cache_size),
            window_counter: Default::default(),
            // 8x: just a heuristic to balance the memory usage and accuracy
            window_limit: cache_size * 8,
            cache_size,
        }
    }

    pub fn new_compact(cache_size: usize) -> Self {
        Self {
            freq_estimator: Estimator::compact(cache_size),
            reuse_estimator: Estimator::compact(cache_size),
            window_counter: Default::default(),
            // 8x: just a heuristic to balance the memory usage and accuracy
            window_limit: cache_size * 8,
            cache_size,
        }
    }

    fn reuse_estimate(&self, freq: u8) -> usize {
        let window_value = self.window_counter.load(Ordering::Relaxed);
        if freq > 0 {
            ((self.window_limit + window_value) as f32) as usize / freq as usize
        } else {
            self.window_limit + window_value
        }
    }
}

pub struct UnoLearner {
    weight_distance: AtomicF32,
    weight_freq: AtomicF32,
    bias: AtomicF32,
    learning_rate: f32,
    pub scale_factor: usize,
}

impl UnoLearner {
    pub fn new(learning_rate: f32, cache_size: usize) -> UnoLearner {
        UnoLearner {
            weight_distance: AtomicF32::new(-0.2),
            weight_freq: AtomicF32::new(0.8),
            bias: AtomicF32::new(0.5),
            learning_rate,
            scale_factor: cache_size,
        }
    }

    // true for freq, false for distance
    pub fn weight_update(&self, freq: u8, distance: usize, touch: bool) {
        let update_value = if touch {
            self.learning_rate
        } else {
            -self.learning_rate
        };

        let distance_scale = (distance / (self.scale_factor + 1)) as f32;
        let weight_distance =
            self.weight_distance.load(Ordering::Relaxed) + update_value * distance_scale;
        let weight_freq = self.weight_freq.load(Ordering::Relaxed) + update_value * freq as f32;
        let bias = self.bias.load(Ordering::Relaxed) + update_value;
        self.weight_distance
            .store(weight_distance, Ordering::Relaxed);
        self.weight_freq.store(weight_freq, Ordering::Relaxed);
        self.bias.store(bias, Ordering::Relaxed);
    }

    pub fn predict(&self, freq: u8, distance: usize) -> f32 {
        let bias = self.bias.load(Ordering::SeqCst);
        let weight_freq = self.weight_freq.load(Ordering::SeqCst);
        let weight_distance = self.weight_distance.load(Ordering::SeqCst);
        let distance_scale = (distance / (self.scale_factor + 1)) as f32;

        bias + weight_freq * freq as f32 + weight_distance * distance_scale
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_uno() {
        let uno = UnoSketch::new(1);
        assert_eq!(uno.get(1), (0, 8));
        assert_eq!(uno.incr(1), (1, 9));
        assert_eq!(uno.incr(1), (2, 5));
        assert_eq!(uno.get(1), (2, 5));

        assert_eq!(uno.get(2), (0, 10));
        assert_eq!(uno.incr(2), (1, 11));
        assert_eq!(uno.incr(2), (2, 6));
        assert_eq!(uno.get(2), (2, 6));

        assert_eq!(uno.incr(3), (1, 13));
        assert_eq!(uno.incr(3), (2, 7));
        assert_eq!(uno.incr(3), (3, 5));
        assert_eq!(uno.incr(3), (4, 4));

        // 8 incr(), now reset

        assert_eq!(uno.incr(3), (4, 2));
        assert_eq!(uno.incr(1), (3, 4));
        assert_eq!(uno.incr(2), (3, 5));
        assert_eq!(uno.get(3), (3, 3));
    }
}