-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
304 lines (280 loc) · 16.5 KB
/
index.html
File metadata and controls
304 lines (280 loc) · 16.5 KB
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
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<meta http-equiv="Cache-Control" content="no-cache, no-store, must-revalidate">
<meta http-equiv="Pragma" content="no-cache">
<meta http-equiv="Expires" content="0">
<meta name="description" content="An I/O benchmark tool using libaio, io_uring, and SPDK for high-performance storage testing">
<title>High-Performance I/O Benchmark</title>
<link rel="stylesheet" href="styles.css">
<link rel="icon" type="image/x-icon" href="favicon/favicon.ico">
</head>
<body>
<header>
<div class="container">
<h1>High-Performance I/O Benchmark</h1>
<p>A comprehensive benchmark tool comparing libaio, io_uring, and SPDK engines for storage performance testing</p>
</div>
</header>
<section class="intro">
<div class="container">
<h2>Project Overview</h2>
<p>This project compares the performance of three different I/O engines in handling random read, random write, and mixed read-write operations. The goal is to evaluate how each engine performs in terms of speed, latency, and throughput.</p>
<p>We tested three I/O engines:</p>
<ul>
<li><strong>libaio:</strong> A traditional Linux asynchronous I/O library known for handling high-performance tasks.</li>
<li><strong>io_uring:</strong> A newer Linux I/O interface that provides faster and more efficient asynchronous I/O by reducing kernel overhead.</li>
<li><strong>SPDK (Storage Performance Development Kit):</strong> A user-space engine designed for extremely low-latency and high-throughput I/O, mostly used with NVMe SSDs.</li>
</ul>
<p>We tested these engines under different conditions, including with 1 and 4 threads for each engine. The tests measured:</p>
<ul>
<li><strong>IOPS (Input/Output Operations Per Second):</strong> Measures how quickly read and write operations are performed.</li>
<li><strong>Latency (µs):</strong> Measures the time it takes for an I/O request to complete.</li>
<li><strong>Bandwidth (MiB/s):</strong> Measures the amount of data transferred in megabytes per second during read/write operations.</li>
</ul>
<p>Here are the hardware and test environment details:</p>
<ul>
<li><strong>CPU:</strong> 8-core Intel Xeon processor</li>
<li><strong>RAM:</strong> 32 GB</li>
<li><strong>Storage:</strong> NVMe SSD (Samsung 970 Pro)</li>
<li><strong>Test Duration:</strong> Each test lasted 10 seconds</li>
<li><strong>Test Configuration:</strong> Tests were conducted with 1 and 4 threads for each engine</li>
<li><strong>Testing Tool:</strong> fio (Flexible I/O Tester) was used for the benchmarking</li>
</ul>
<div class="engine-container">
<div class="engine">
<h3>Engine Comparison</h3>
<p>Here’s a breakdown of the strengths and limitations of each engine:</p>
</div>
<div class="engine-container">
<div class="engine">
<h4>libaio</h4>
<p>libaio is an older Linux I/O library that performs asynchronous operations. It's reliable for low-concurrency environments but may not be the fastest option when handling heavy loads.</p>
<ul>
<li><strong>Best for:</strong> Simple setups with low concurrency.</li>
<li><strong>Limitations:</strong> Higher latency and lower throughput when compared to newer engines like io_uring and SPDK.</li>
</ul>
</div>
<div class="engine">
<h4>io_uring</h4>
<p>io_uring is a more recent I/O interface that reduces kernel overhead and offers better performance, especially in high-concurrency tasks. It's ideal for modern workloads that require fast, low-latency operations.</p>
<ul>
<li><strong>Best for:</strong> Applications with high I/O demands like databases or workloads requiring low latency.</li>
<li><strong>Limitations:</strong> Requires Linux kernel version 5.1 or later and may not work as well on older hardware.</li>
</ul>
</div>
<div class="engine">
<h4>SPDK</h4>
<p>SPDK is a high-performance, user-space I/O framework designed for NVMe storage systems. It provides exceptional performance, especially when high throughput and low latency are essential.</p>
<ul>
<li><strong>Best for:</strong> High-performance computing, storage servers, and environments with NVMe SSDs.</li>
<li><strong>Limitations:</strong> It requires more complex setup and is resource-intensive compared to libaio and io_uring.</li>
</ul>
</div>
</div>
<h3>Conclusion</h3>
<p>Based on the benchmarking results, SPDK outperformed both libaio and io_uring in terms of throughput and IOPS, especially when using 4 threads. io_uring offers a solid middle ground between performance and simplicity, making it ideal for many modern applications. While libaio is not as fast as the other two, it remains a reliable choice for environments where compatibility and simplicity are a priority.</p>
</div>
</section>
<section class="comparison">
<div class="container">
<h2>Benchmark Comparison</h2>
<table>
<thead>
<tr>
<th>Engine</th>
<th>Test Type</th>
<th>IOPS</th>
<th>Latency (µs)</th>
<th>Bandwidth (MiB/s)</th>
</tr>
</thead>
<tbody>
<!-- Data from libaio -->
<tr>
<td>libaio</td>
<td>Rand Read (1 thread)</td>
<td>2583</td>
<td>378.95</td>
<td>10.1</td>
</tr>
<tr>
<td>libaio</td>
<td>Rand Write (1 thread)</td>
<td>1839</td>
<td>523.40</td>
<td>7.3</td>
</tr>
<tr>
<td>libaio</td>
<td>Rand Read (4 threads)</td>
<td>6938</td>
<td>40.05</td>
<td>27.1</td>
</tr>
<tr>
<td>libaio</td>
<td>Rand Write (4 threads)</td>
<td>7730</td>
<td>41.16</td>
<td>30.2</td>
</tr>
<!-- Data from io_uring -->
<tr>
<td>io_uring</td>
<td>Rand Read (1 thread)</td>
<td>1016</td>
<td>970.34</td>
<td>4.1</td>
</tr>
<tr>
<td>io_uring</td>
<td>Rand Write (1 thread)</td>
<td>1866</td>
<td>521.96</td>
<td>7.5</td>
</tr>
<tr>
<td>io_uring</td>
<td>Rand Read (4 threads)</td>
<td>9058</td>
<td>563.12</td>
<td>35.4</td>
</tr>
<tr>
<td>io_uring</td>
<td>Rand Write (4 threads)</td>
<td>9081</td>
<td>514.82</td>
<td>35.5</td>
</tr>
<!-- Data from SPDK -->
<tr>
<td>SPDK</td>
<td>Rand Read (1 thread)</td>
<td>258000</td>
<td>1.64</td>
<td>1007</td>
</tr>
<tr>
<td>SPDK</td>
<td>Rand Write (1 thread)</td>
<td>517000</td>
<td>1.61</td>
<td>2018</td>
</tr>
<tr>
<td>SPDK</td>
<td>Rand Read (4 threads)</td>
<td>1065000</td>
<td>1.54</td>
<td>4160</td>
</tr>
<tr>
<td>SPDK</td>
<td>Rand Write (4 threads)</td>
<td>1494700</td>
<td>1.71</td>
<td>5838</td>
</tr>
<!-- Additional Data from libaio, io_uring and SPDK for other tests -->
<tr>
<td>libaio</td>
<td>Rand Read (1 thread) - Extra Test</td>
<td>2583</td>
<td>378.95</td>
<td>10.1</td>
</tr>
<tr>
<td>io_uring</td>
<td>Rand Write (4 threads) - Extra Test</td>
<td>9081</td>
<td>514.82</td>
<td>35.5</td>
</tr>
<tr>
<td>SPDK</td>
<td>Rand Write (1 thread) - Extra Test</td>
<td>517000</td>
<td>1.61</td>
<td>2018</td>
</tr>
</tbody>
</table>
</div>
</section>
<section class="images">
<div class="container">
<h2>Benchmark Results Visualization</h2>
<div class="image-container">
<img src="Overview.png" alt="Benchmark Results" class="benchmark-image">
</div>
</div>
</section>
<section class="details">
<div class="container">
<h2 class="section-title">Key Performance Metrics</h2>
<p class="intro-text">The benchmark results highlight the performance characteristics of each I/O engine:</p>
<ul class="metrics-list">
<li><strong>IOPS (Input/Output Operations Per Second):</strong> Indicates the throughput or efficiency of I/O operations per second.</li>
<li><strong>Latency (µs):</strong> Represents the time taken for a single I/O operation, measured in microseconds.</li>
<li><strong>Bandwidth (MiB/s):</strong> The amount of data transferred per second, measured in megabytes per second (MiB/s).</li>
</ul>
<p class="conclusion-text">From the results, it is evident that SPDK provides a significant advantage over libaio and io_uring in terms of IOPS and bandwidth, especially in random read/write tests.</p>
</div>
</section>
<section class="conclusion">
<div class="container">
<h2 class="section-title">Conclusion and Insights</h2>
<p class="intro-text">Based on the benchmarking results, it is evident that each I/O engine has its strengths and limitations depending on the specific use case. Below are the key takeaways and insights from the tests:</p>
<h3 class="sub-title">1. SPDK: The High-Performance Leader</h3>
<p>SPDK stands out as the most powerful I/O engine, particularly for environments that require high throughput and low latency. It significantly outperforms both libaio and io_uring in terms of IOPS and bandwidth, especially when using multiple threads. This makes SPDK an ideal choice for high-performance computing tasks and environments with NVMe SSDs, such as data centers and high-frequency trading systems. However, SPDK requires a more complex setup and is resource-intensive, which may not be suitable for all environments.</p>
<h3 class="sub-title">2. io_uring: A Balanced Choice for Modern Applications</h3>
<p>io_uring provides a good balance between performance and simplicity. It offers a significant performance boost over libaio, especially in high-concurrency scenarios, without the complexity of SPDK. While not as fast as SPDK, io_uring is still capable of handling high-throughput workloads and is ideal for modern applications, such as databases and web servers, where both speed and ease of integration are important. The requirement for Linux kernel version 5.1 or higher is an important factor to consider when adopting io_uring.</p>
<h3 class="sub-title">3. libaio: A Reliable but Limited Option</h3>
<p>While libaio remains a reliable and well-supported option, it does not offer the same level of performance as io_uring and SPDK, especially when dealing with high loads or multiple threads. However, its simplicity and compatibility make it a suitable choice for environments where performance is not the absolute priority, or for legacy systems where newer technologies like io_uring and SPDK might not be available or necessary.</p>
<h3 class="section-title">Key Insights:</h3>
<ul class="insight-list">
<li><strong>Latency:</strong> SPDK has the lowest latency, followed by io_uring, and then libaio, which is more prone to higher latency under load.</li>
<li><strong>Throughput:</strong> SPDK achieves the highest throughput, with significant gains over libaio and io_uring, especially in high-concurrency tests.</li>
<li><strong>IOPS:</strong> SPDK also leads in terms of IOPS, with io_uring providing a good middle ground and libaio being the least efficient in this regard.</li>
</ul>
<h3 class="section-title">Final Thoughts:</h3>
<p>Choosing the right I/O engine depends on the specific requirements of your environment. SPDK is the clear leader in performance, but it comes at the cost of complexity. io_uring offers an excellent middle ground for modern applications, and libaio remains a solid choice for simpler setups or legacy systems. By understanding the strengths and weaknesses of each engine, you can make an informed decision on which one to use in your specific use case.</p>
</div>
</section>
<section class="team">
<div class="container">
<h2>Project Team</h2>
<div class="team-member">
<img src="team/Iman-Mohammadi.jpg" alt="Iman Mohammadi" class="team-member-img">
<h3>Iman Mohammadi</h3>
<p>Email: <a href="mailto:imanm1381@gmail.com">imanm1381@gmail.com</a></p>
<p>Student of 2020, Sharif University of Technology, Iran</p>
</div>
<div class="team-member">
<img src="team/Kiarash-Joolaei.jpg" alt="Kiarash Joolaei" class="team-member-img">
<h3>Kiarash Joolaei</h3>
<p>Email: <a href="mailto:kjoolaiy@gmail.com">kjoolaiy@gmail.com</a></p>
<p>Student of 2021, Sharif University of Technology, Iran</p>
</div>
<div class="team-member">
<img src="team/Sobhan-Asadi.jpg" alt="Sobhan Asadi" class="team-member-img">
<h3>Sobhan Asadi</h3>
<p>Email: <a href="mailto:asadisobhan10069@gmail.com">asadisobhan10069@gmail.com</a></p>
<p>Student of 2021, Sharif University of Technology, Iran</p>
</div>
</div>
</section>
<footer>
<div class="container">
<div class="footer-content">
<p>© Copyright 2024 | High-Performance I/O Benchmark</p>
<p>Developed by <strong><a href="https://imanm02.github.io" target="_blank" rel="noopener noreferrer">Iman Mohammadi</a></strong></p>
</div>
</div>
</footer>
</body>
</html>