caidespries.github.io/project2.html at main · CaideSpries/caidespries.github.io · GitHub

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
<!DOCTYPE html>
<html lang="en">

<head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <title>Comparative Analysis of MD5 & MD6 Hashing Implementations</title>
    <link rel="stylesheet" href="assets/css/projects.css">
</head>

<body>
<div class="container">
    <header>
        <h1>Comparative Analysis of MD5 & MD6 Hashing Implementations</h1>
    </header>
    <main>
        <section id="project-overview">
            <h2>Project Overview</h2>
            <div style="text-align: center; max-width: 90%; margin: 1rem auto;">
                <p>Hashing algorithms are fundamental to data integrity and security, but not all are created equal.
                    This project explores the critical trade-offs between the classic, sequential MD5 algorithm and the
                    modern, parallel-by-design MD6 algorithm. We conducted a comparative analysis by implementing these
                    hashes across three distinct platforms:</p>
            </div>
            <ul>
                <li><strong>Standard Software (C++):</strong> To establish a performance baseline.</li>
                <li><strong>Hardware Acceleration (Verilog on an FPGA):</strong> To test the limits of raw speed
                    with dedicated hardware.
                </li>
                <li><strong>GPU Parallel Computing (OpenCL):</strong> To investigate how a parallel framework
                    handles both sequential and parallel-native algorithms.
                </li>
            </ul>
            <div style="text-align: center; max-width: 90%; margin: 1rem auto;">
                <p>The core objective was to quantify the performance differences in execution time and throughput,
                    revealing the optimal use case for each approach.</p>
            </div>
        </section>
        <section id="algorithms-explained">
            <h2>The Algorithms: MD5 vs. MD6</h2>
            <div style="text-align: center; max-width: 90%; margin: 1rem auto;">
                <p>The key difference between the two algorithms lies in their structure. MD5 processes data in a
                    strictly
                    sequential order, where each step depends on the previous one. This makes it inherently difficult to
                    parallelise. The MD6 algorithm, in contrast, was designed with a tree-like structure that allows
                    large
                    files to be broken down and hashed concurrently, making it ideal for modern multi-core processors
                    and
                    parallel environments.</p>
            </div>

            <figure style="text-align: center;">
                <img src="assets/img/Projects Docs/project2/flowchart-md5-hashing-algorithm.webp"
                     alt="Flowchart showing the sequential nature of the MD5 hashing algorithm"
                     style="max-width:100%; height:auto;">
                <figcaption>Figure 1: MD5 Algorithm Flowchart illustrating its sequential processing structure
                </figcaption>
            </figure>
        </section>
        <section id="methodology">
            <h2>Methodology: Implementations Across Platforms</h2>
            <div style="text-align: center; max-width: 90%; margin: 1rem auto;">
                <p>To perform a comprehensive comparison, we developed several implementations:</p>
            </div>
            <ul>
                <li><strong>C++ (MD5 & MD6):</strong> Served as our "golden standard" for software performance. We
                    created both sequential and parallel (multi-threaded) versions of MD6 to directly measure the
                    benefits of parallelization in software.
                </li>
                <li><strong>Verilog (MD5):</strong> A hardware implementation designed for a Nexys A7 FPGA. The goal was
                    to offload the entire hashing computation to a dedicated chip to achieve the lowest possible
                    execution time.
                </li>
                <li><strong>OpenCL (MD5):</strong> A GPU-based implementation to test whether the massively parallel
                    architecture of a GPU could overcome the sequential nature of the MD5 algorithm.
                </li>
            </ul>
        </section>
        <section id="project-results">
            <h2>Results and Performance Analysis</h2>
            <div style="text-align: center; max-width: 90%; margin: 1rem auto;">
                <p>Our experiments, which tested files of various sizes, revealed significant performance differences
                    across the implementations. The two graphs below summarise the overall comparison of execution time
                    and throughput for the software-based approaches.</p>
            </div>

            <figure style="text-align: center;">
                <img src="assets/img/Projects Docs/project2/execution-time-across-implementations.webp"
                     alt="Line graph comparing execution time across different implementations including C++ MD5, C++ MD6, Parallel MD6, and OpenCL MD5"
                     style="max-width:100%; height:auto;">
                <figcaption>Figure 2: Comparison of Execution Time Across Implementations</figcaption>
            </figure>

            <figure style="text-align: center;">
                <img src="assets/img/Projects Docs/project2/throughput-comparisons.webp"
                     alt="Line graph comparing throughput in MB/s across different implementations including C++ MD5, C++ MD6, Parallel MD6, and OpenCL MD5"
                     style="max-width:100%; height:auto;">
                <figcaption>Figure 3: Comparison of Throughput Across Implementations</figcaption>
            </figure>

            <div style="text-align: center; max-width: 90%; margin: 1rem auto;">
                <p>As the graphs clearly show, the parallelised C++ MD6 implementation is the undisputed winner for
                    handling large data volumes. It consistently demonstrates the lowest execution time and the highest
                    throughput, reaching nearly 500 MB/s. This directly confirms the architectural advantages of a
                    parallel-aware algorithm.</p>

                <h3>Key Performance Highlights</h3>

                <p>Digging deeper into the data reveals two fascinating edge cases:</p>
            </div>

            <ul>
                <li><strong>The Power of Dedicated Hardware:</strong> For very small messages (0 bytes), the Verilog
                    FPGA implementation was astonishingly fast, achieving a speedup of over 132x compared to the
                    standard C++ MD5. Its execution time was a mere 8.88 nanoseconds. This highlights the incredible
                    efficiency of hardware acceleration for highly specific, repetitive tasks, even though its input
                    size was limited.
                </li>
                <li><strong>An Insightful Failure:</strong> The OpenCL implementation of MD5 was significantly slower
                    than the basic C++ version. This is a critical finding: applying a parallel framework to an
                    inherently sequential algorithm does not guarantee a speedup. The overhead of transferring data to
                    the GPU and managing the kernel outweighed any potential benefits, proving that the algorithm's
                    design is paramount.
                </li>
            </ul>
        </section>
        <section id="acceptance-testing">
            <h2>Acceptance Testing: Proof of Correctness</h2>
            <div style="text-align: center; max-width: 90%; margin: 1rem auto;">
                <p>Beyond performance, we rigorously tested each implementation to ensure it produced correct,
                    standard-compliant hash values. All implementations passed verification against known test vectors
                    for both MD5 and MD6.</p>
            </div>

            <div class="image-grid"
                 style="display: flex; flex-wrap: wrap; justify-content: center; gap: 1rem; margin-top: 2rem;">
                <figure style="text-align: center; flex: 1; min-width: 45%;">
                    <img src="assets/img/Projects Docs/project2/sequential-md5-base-test.webp"
                         alt="Terminal output showing successful verification of sequential MD5 implementation"
                         style="max-width:100%; height:auto;">
                    <figcaption>Figure 4: Sequential MD5 Implementation Verification</figcaption>
                </figure>

                <figure style="text-align: center; flex: 1; min-width: 45%;">
                    <img src="assets/img/Projects Docs/project2/parallel-md5-accuracy-test.webp"
                         alt="Terminal output showing successful verification of parallel MD5 implementation"
                         style="max-width:100%; height:auto;">
                    <figcaption>Figure 5: Parallel MD5 Implementation Verification</figcaption>
                </figure>

                <figure style="text-align: center; flex: 1; min-width: 45%;">
                    <img src="assets/img/Projects Docs/project2/md6-length-accuracy-test.webp"
                         alt="Terminal output showing successful verification of MD6 implementation with different input lengths"
                         style="max-width:100%; height:auto;">
                    <figcaption>Figure 6: MD6 Implementation Verification</figcaption>
                </figure>

                <figure style="text-align: center; flex: 1; min-width: 45%;">
                    <img src="assets/img/Projects Docs/project2/verilog-accuracy-test.webp"
                         alt="Output showing successful verification of Verilog FPGA implementation"
                         style="max-width:100%; height:auto;">
                    <figcaption>Figure 7: Verilog FPGA Implementation Verification</figcaption>
                </figure>
            </div>
        </section>
        <section id="conclusion">
            <h2>Conclusion</h2>
            <div style="text-align: center; max-width: 90%; margin: 1rem auto;">
                <p>This project successfully quantified the trade-offs between software, GPU, and hardware
                    implementations of sequential and parallel hashing algorithms. The key findings are clear:</p>
            </div>
            <ul>
                <li>For high-throughput applications on modern systems, the parallelised MD6 algorithm is the superior
                    choice.
                </li>
                <li>For specialised, low-latency tasks with small inputs, a hardware (FPGA) implementation offers
                    unparalleled speed.
                </li>
                <li>A parallel architecture like a GPU cannot fix a fundamentally sequential algorithm like MD5; in
                    fact, the overhead can make performance worse.
                </li>
            </ul>
            <div style="text-align: center; max-width: 90%; margin: 1rem auto;">
                <p>Ultimately, the analysis provides valuable insights into selecting the right tool for the job,
                    demonstrating the crucial interplay between algorithm design and the underlying hardware
                    platform.</p>
            </div>
        </section>
        <section id="project-document">
            <h2>Project Document</h2>
            <div style="text-align: center; max-width: 90%; margin: 1rem auto;">
                <p>For more detailed information, you can access the following resources:</p>
                <div style="display: flex; justify-content: center; gap: 20px; flex-wrap: wrap; margin-top: 1rem;">
                    <a href="assets/docs/Projects Docs/project2/Analysis%20of%20Hashing%20Algorithms%20on%20FPGAs.pdf"
                       target="_blank"
                       class="button-link"
                       style="display: inline-block; padding: 10px 15px; background-color: #3498db; color: white; text-decoration: none; border-radius: 4px;">
                        View Project Document (PDF)
                    </a>
                    <a href="https://github.com/CaideSpries/EEE4120F-YODA"
                       target="_blank"
                       class="button-link"
                       style="display: inline-block; padding: 10px 15px; background-color: #333; color: white; text-decoration: none; border-radius: 4px;">
                        View Source Code on GitHub
                    </a>
                </div>
            </div>
        </section>
    </main>
    <footer>
        <a href="index.html">Back to Homepage</a>
    </footer>
</div>
</body>

</html>