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
|
/**
* Copyright (C) 2023-present MongoDB, Inc.
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the Server Side Public License, version 1,
* as published by MongoDB, Inc.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* Server Side Public License for more details.
*
* You should have received a copy of the Server Side Public License
* along with this program. If not, see
* <http://www.mongodb.com/licensing/server-side-public-license>.
*
* As a special exception, the copyright holders give permission to link the
* code of portions of this program with the OpenSSL library under certain
* conditions as described in each individual source file and distribute
* linked combinations including the program with the OpenSSL library. You
* must comply with the Server Side Public License in all respects for
* all of the code used other than as permitted herein. If you modify file(s)
* with this exception, you may extend this exception to your version of the
* file(s), but you are not obligated to do so. If you do not wish to do so,
* delete this exception statement from your version. If you delete this
* exception statement from all source files in the program, then also delete
* it in the license file.
*/
#include "mongo/tools/workload_simulation/workload_characteristics.h"
#include <cmath>
namespace mongo::workload_simulation {
namespace {
int32_t independentThroughput(int32_t throughputAtOptimalConcurrency,
int32_t optimalConcurrency,
int32_t concurrency) {
if (concurrency == optimalConcurrency) {
return throughputAtOptimalConcurrency;
}
// Modeling throughput as piecewise parabolic functions in the regions of interest, flat
// past 5x optimal. The 5x threshold is somewhat arbitrary, but is chosen so that the
// degradation in performance slightly past the optimal is significant, but we do not
// converge on the minimum immediately.
if (concurrency < optimalConcurrency) {
// Fit parabola to (0, 0), (optimalConcurrency, throughputAtOptimalConcurrency),
// (2 * optimalConcurrency, 0).
const double a = static_cast<double>(throughputAtOptimalConcurrency) * (-1.0) /
(optimalConcurrency * optimalConcurrency);
const double b = a * optimalConcurrency * -2.0;
return (a * concurrency * concurrency) + (b * concurrency);
} else if (concurrency < 5 * optimalConcurrency) {
// Fit parabola to (optimalConcurrency, throughputAtOptimalConcurrency),
// (5 * optimalConcurrency, 0.1 * throughputAtOptimalConcurrency),
// (9 * optimalConcurrency, throughputAtOptimalConcurrency).
const double a = (static_cast<double>(throughputAtOptimalConcurrency) * 9.0) /
(160.0 * optimalConcurrency * optimalConcurrency);
const double b = ((static_cast<double>(throughputAtOptimalConcurrency) * -9.0) /
(40.0 * optimalConcurrency)) -
(6.0 * a * optimalConcurrency);
const double c = static_cast<double>(throughputAtOptimalConcurrency) -
(a * optimalConcurrency * optimalConcurrency) - (b * optimalConcurrency);
return static_cast<int32_t>((a * concurrency * concurrency) + (b * concurrency) + c);
}
return throughputAtOptimalConcurrency / 10;
}
} // namespace
MockWorkloadCharacteristics::MockWorkloadCharacteristics(RWPair optimalConcurrency,
RWPair throughputAtOptimalConcurrency,
double jitterDev)
: _optimalConcurrency(optimalConcurrency),
_throughputAtOptimalConcurrency(throughputAtOptimalConcurrency),
_jitterDev{jitterDev},
// If the deviation is 0, we will always return 0, so we don't care about the distribution.
// However, it's not valid to pass 0 as the sigma for a normal distribution, so we'll need to
// override it.
_jitterDist{0.0, _jitterDev == 0.0 ? 1.0 : _jitterDev} {
invariant(_jitterDev >= 0.0);
invariant(_jitterDev < 1.0);
}
MockWorkloadCharacteristics::~MockWorkloadCharacteristics() {}
int32_t MockWorkloadCharacteristics::readThroughput(RWPair concurrency) const {
return _throughput(concurrency).read;
}
int32_t MockWorkloadCharacteristics::writeThroughput(RWPair concurrency) const {
return _throughput(concurrency).write;
}
Nanoseconds MockWorkloadCharacteristics::readLatency(RWPair concurrency) const {
return _latency(concurrency.read, readThroughput(concurrency));
}
Nanoseconds MockWorkloadCharacteristics::writeLatency(RWPair concurrency) const {
return _latency(concurrency.write, writeThroughput(concurrency));
}
RWPair MockWorkloadCharacteristics::optimal() const {
return _optimalConcurrency.load();
}
void MockWorkloadCharacteristics::reset(RWPair newOptimalConcurrency, RWPair newOptimalThroughput) {
_optimalConcurrency.store(newOptimalConcurrency);
_throughputAtOptimalConcurrency.store(newOptimalThroughput);
}
double MockWorkloadCharacteristics::_jitter() const {
if (_jitterDev == 0.0) {
return 0.0;
}
stdx::lock_guard lk{_mutex};
return _jitterDist(_rng);
}
Nanoseconds MockWorkloadCharacteristics::_latency(int32_t concurrency, int32_t throughput) const {
invariant(throughput != 0);
invariant(throughput < 1'000'000);
return Nanoseconds{static_cast<int64_t>(static_cast<double>(1'000'000'000) * concurrency *
std::clamp(1.0 + _jitter(), 0.0, kMaxJitterMultiplier) /
throughput)};
}
ParabolicWorkloadCharacteristics::ParabolicWorkloadCharacteristics(
RWPair optimalConcurrency, RWPair throughputAtOptimalConcurrency, double jitterDev)
: MockWorkloadCharacteristics(optimalConcurrency, throughputAtOptimalConcurrency, jitterDev) {}
RWPair ParabolicWorkloadCharacteristics::_throughput(RWPair concurrency) const {
RWPair throughputAtOptimalConcurrency = _throughputAtOptimalConcurrency.load();
RWPair optimalConcurrency = _optimalConcurrency.load();
RWPair independentThroughputs = {
independentThroughput(
throughputAtOptimalConcurrency.read, optimalConcurrency.read, concurrency.read),
independentThroughput(
throughputAtOptimalConcurrency.write, optimalConcurrency.write, concurrency.write)};
double optimalRatio =
static_cast<double>(optimalConcurrency.read) / std::max(1, optimalConcurrency.write);
double ratio = static_cast<double>(concurrency.read) / std::max(1, concurrency.write);
RWPair scaledThroughputs = independentThroughputs;
if (ratio > optimalRatio) {
// Too many reads, penalize writes
scaledThroughputs.write *= std::sqrt(optimalRatio / ratio);
} else {
// Too many writes, penalize reads
scaledThroughputs.read *= std::sqrt(ratio / optimalRatio);
}
return scaledThroughputs;
}
} // namespace mongo::workload_simulation
|
