##plugins.themes.bootstrap3.article.main##

This study investigates the impact of ad-blockers on system power consumption in a computing environment equipped with an AI accelerator. The increasing prevalence of online advertisements has raised concerns about system performance and energy efficiency, prompting many users to turn to ad-blockers. However, the effectiveness of ad-blockers on power consumption, especially in systems equipped with specialized AI accelerators, remains underexplored. In this research, we evaluate the power usage, GPU utilization, and memory consumption of computers running ad-blockers on both Windows and Ubuntu operating systems. The study compared traditional CPU/GPU methods with AI-accelerated scenarios, using popular ad-blockers such as AdBlock, Adblock Plus, uBlock, uBlock Origin, and uBlock Origin Lite. Results indicate that uBlock Origin and uBlock Origin Lite were the most efficient, significantly reducing power consumption and memory usage compared to other ad-blockers. However, multimedia-heavy websites presented challenges, with increased resource usage observed. The findings emphasize the importance of choosing appropriate ad-blockers to enhance energy efficiency, optimize system resources, and contribute to sustainable computing.

Introduction

In the digital age, online advertising has become an integral part of the Internet ecosystem, generating significant revenue for content creators and service providers. However, the proliferation of online advertisements also presents various challenges for end-users, including privacy concerns, security risks, distraction, and diminished browsing experiences. Moreover, the increasing volume of advertisements has a direct impact on computer performance and energy consumption, as ads require additional computing power and data to render. Consequently, users have turned to ad-blockers as a solution to mitigate these issues and enhance their browsing experience by eliminating intrusive ads.

Ad-blockers are software tools designed to detect and prevent advertisements from loading on web pages, thereby reducing network bandwidth usage, enhancing browsing speed, and addressing privacy concerns. Despite these benefits, the operation of ad-blockers itself consumes system resources, such as CPU, GPU, and memory, contributing to increased power consumption. Understanding the implications of ad-blockers on system energy consumption is essential, especially in a world increasingly focused on environmental sustainability and reducing carbon footprints. As laptops, desktops, and mobile devices are widely used, optimizing their energy efficiency not only extends battery life but also minimizes the environmental impact of technology use.

Recent advancements in computer hardware have introduced specialized components, such as AI accelerators, designed to optimize specific computational tasks more efficiently than traditional CPUs and GPUs. AI accelerators, such as Tensor Processing Units (TPUs) and dedicated neural processing units (NPUs), leverage machine learning capabilities to handle complex and repetitive tasks, providing higher performance while consuming less power. These accelerators have shown significant potential in enhancing the efficiency of various workloads, including image recognition, natural language processing, and even content filtering. This raises an intriguing question: Can AI accelerators be leveraged to make ad-blocking more energy efficient?

This study aims to explore the power consumption of computers running ad-blockers on systems equipped with AI accelerators. Specifically, it investigates whether the use of AI acceleration can effectively reduce the energy footprint associated with ad blocking compared to conventional methods that rely solely on CPU or GPU resources. By examining power consumption under different scenarios—such as using traditional ad-blockers versus AI-optimized ad-blocking mechanisms—this research aims to provide valuable insights into optimizing power usage for enhanced sustainability.

Furthermore, this research considers the broader implications of power consumption in modern computing environments. As consumers increasingly rely on mobile devices and laptops, battery life has become a critical factor in determining user satisfaction. Ad-blockers, while improving browsing speed and privacy, might paradoxically drain more power if not implemented efficiently. Therefore, understanding how AI accelerators impact the energy usage of ad-blocking applications can lead to the development of more energy-efficient solutions, benefiting both individual users and contributing to a more sustainable technology landscape.

The results of this study will be useful for multiple stakeholders, including end-users seeking to maximize device battery life, developers aiming to create efficient ad-blocking solutions, and researchers focusing on sustainable computing practices. The research also addresses the importance of energy efficiency in the design and implementation of privacy-enhancing technologies, highlighting how hardware advancements can be utilized to balance privacy, performance, and sustainability in an increasingly connected world.

Literature Review

The rapid advancement of artificial intelligence (AI) has led to an increasing demand for efficient hardware accelerators capable of handling complex computations. Traditional processors struggle to meet the performance and energy efficiency requirements of modern AI applications. As a result, specialized AI hardware accelerators have emerged as critical components in the deployment of AI systems. A comprehensive overview of AI hardware accelerators, including their architectural designs, performance metrics, and integration challenges, is provided by Mishra et al. [1]. This foundational knowledge sets the stage for exploring specific advancements in energy-efficient accelerator designs and their applications across various domains.

Energy efficiency is a paramount concern in the design of deep learning accelerators, especially given the growing complexity of neural network models. The Eyeriss accelerator introduced by Chen et al. [2] is an energy-efficient reconfigurable hardware designed for deep convolutional neural networks (CNNs). Eyeriss employs a novel dataflow called Row Stationary, which optimizes data reuse and minimizes data movement, leading to significant energy savings.

Building on the need for energy efficiency in personalized applications, an accelerator facilitating in situ personalization on smart devices is proposed by Choi et al. [3]. This design allows for on-device training of deep CNNs, reducing dependency on cloud services and enhancing user privacy. The accelerator optimizes energy consumption by leveraging low-precision computations and efficient memory hierarchies.

In the quest for adaptability, the UNPU accelerator presented by Lee et al. [4] supports fully variable weight bit precision. UNPU dynamically adjusts the precision of weights during neural network operations, balancing the trade-off between energy efficiency and computational accuracy. This flexibility enables the accelerator to cater to a wide range of applications with varying precision requirements.

The separation of memory and processing units in traditional architectures leads to significant energy consumption due to data movement. In-memory computing addresses this challenge by performing computations within the memory elements themselves. The Pipelayer accelerator introduced in Song et al. [5] leverages resistive random-access memory (ReRAM) for deep learning applications. By minimizing data transfer between memory and processing units and supporting pipeline parallelism, Pipelayer reduces energy consumption and enhances throughput.

The challenges and solutions associated with in-memory computing for AI accelerators are extensively discussed in Cherupally et al. [6]. Limitations of existing memory technologies are examined, and architectural innovations are proposed to overcome them. This work emphasizes the importance of co-designing hardware and algorithms to fully exploit the benefits of in-memory computing, particularly in terms of energy efficiency and performance scalability.

To accommodate the diverse computational patterns of deep learning models, flexibility in dataflow mapping is essential. The MAERI architecture presented in Kwon et al. [7] enables flexible dataflow mapping over deep neural network (DNN) accelerators through reconfigurable interconnects. MAERI’s design allows for the efficient execution of various neural network layers by adapting the dataflow to the specific requirements of each layer. This adaptability enhances both performance and energy efficiency, as the accelerator can optimize resource utilization dynamically.

Understanding the landscape of AI accelerators requires comprehensive benchmarking and analysis. A survey of machine learning accelerators is conducted in [8], examining architectural features, performance metrics, and application domains. This work provides valuable insights into the strengths and weaknesses of various accelerator designs, highlighting trends in the field.

A systematic study on benchmarking AI inference accelerators is presented by Jiang et al. [9], proposing methodologies for evaluating accelerator performance across different workloads and models. The importance of standardized benchmarks for fair comparisons is emphasized.

Significant surveys contributed by Capra et al. [10], [11] update the state of efficient hardware architectures for accelerating deep CNNs and discuss hardware and software optimizations for accelerating deep neural networks. These works collectively offer a thorough understanding of the current state and evolution of AI hardware accelerators.

The deployment of AI accelerators extends beyond traditional computing applications into specialized fields like healthcare and biomedical engineering. Hardware implementations of deep network accelerators tailored for these applications are explored in Azghadi et al. [12]. This research highlights the unique challenges in these domains, such as the need for high precision and reliability and how customized hardware solutions can address these requirements while maintaining energy efficiency.

AI technologies play a significant role in optimizing energy consumption in smart devices and systems. A review of AI-empowered methods for smart energy consumption is provided in Himeur et al. [13], focusing on load forecasting, anomaly detection, and demand response. The work discusses how AI algorithms, when implemented efficiently on hardware accelerators, can lead to substantial energy savings in smart grids and buildings.

The complexity of modern chip design necessitates innovative methodologies to accelerate development cycles. A graph placement methodology leveraging machine learning to expedite chip design processes is introduced in Mirhoseini et al. [14]. By formulating chip design as a reinforcement learning problem, this approach achieves superior placement quality in a fraction of the time required by traditional methods. This advancement not only speeds up the development of AI accelerators but also contributes to their performance and energy efficiency.

Personalized recommendation systems are computationally intensive and require efficient processing of large datasets. The RecNMP architecture proposed by Ke et al. [15] accelerates personalized recommendation tasks using near-memory processing. By bringing computation closer to memory, RecNMP reduces data movement overhead and improves energy efficiency. This approach demonstrates the applicability of AI accelerators in enhancing user-centric services while managing energy consumption effectively.

Ad-blockers have become prevalent tools for users seeking to enhance their browsing experience by eliminating unwanted advertisements. The economic implications of ad-blocker platforms on advertisers and the internet ecosystem are analyzed by Ray et al. [16]. The study highlights the tension between user preferences and the revenue models of content providers.

The effects of ad-blocker adoption on digital piracy are explored by Datta and Madio [17], discussing whether ad-blockers serve as a deterrent to piracy by improving user experience on legitimate platforms or inadvertently encourage piracy by disrupting revenue streams.

A lab experiment assessing the impact of ad-blockers on consumer behavior is conducted by Frik et al. [18]. The findings suggest that while ad-blockers improve user experience, they also alter consumer engagement with content and advertisements.

The nuanced view of ad-blockers being beneficial or detrimental to the digital economy is provided by Aseri et al. [19], exploring the complex interplay between user experience, content monetization, and the sustainability of online services.

Beyond economic implications, ad-blockers have a significant impact on energy consumption and device performance. The contribution of open-source ad-blockers to energy conservation is investigated in Pearce [20]. The study demonstrates that blocking advertisements reduces data usage and processing demands, leading to lower energy consumption on user devices.

Building better mobile web browsers for ad blocking from an energy perspective is the focus of Heitmann et al. [21]. The analysis of different ad-blocking strategies and their effects on the energy efficiency of mobile browsers provides insights into optimizing both user experience and device battery life.

An experimental study on the energy and bandwidth costs of web advertisements on smartphones is conducted in Albasir et al. [22]. The findings reveal that advertisements significantly increase energy consumption and data usage, underscoring the potential benefits of ad-blocking technologies in prolonging battery life and reducing costs for users.

Experimental Design Overview

The primary objective of this experiment was to evaluate the impact of using different ad-blockers on system power consumption, specifically focusing on systems equipped with AI accelerators. The study compared power usage between scenarios where ads were blocked versus scenarios without ad-blockers on both Windows and Ubuntu systems. Additionally, the experiment included comparisons between different ad-blockers across various websites, as well as between the Windows and Ubuntu operating systems. The hypothesis was that using an ad-blocker would reduce overall power consumption by minimizing the processing required for advertisements, particularly when using an AI-accelerated setup.

Hardware and Software Specifications

1. Processor Type: AMD Ryzen 9 4900H with Radeon Graphics, 3.30 GHz

2. AI Accelerator Model: Nvidia GeForce RTX 2060

3. AI Accelerator and Specifications: Built on the 12 nm process and based on the TU106 graphics processor In its TU106-200-KA-A1 variant, the card supports DirectX 12 Ultimate. The Second generation Tenson Cores (succeeding Volta’s) work in cooperation with the RT cores and their AI features

4. System Architecture: 64-bit operating system, x64-based processor

5. Memory Specifications: Installed RAM: 24.0 GB (23.4 GB usable)

6. Network Configuration: Wired network connection, 1.5 Gbps speed

7. Operating System and Version:

• Windows 11 Home, Version 23H2

• Ubuntu (Version 24.04.1)

8. Ad-Blockers Used:

• Tested without an ad-blocker and with the following: AdBlock, AdBlock Plus, uBlock, uBlock Origin, and uBlock Origin Lite.

9. Additional Software/Dependencies:

• Web Browser: Latest version of Google Chrome for both Windows and Ubuntu systems.

Measurement Tools

1. Power Measurement Tool Method:

• For Windows, HWinfo was used to monitor both CPU and GPU power consumption.

• For Ubuntu, a custom bash script was utilized to log CPU and GPU power, as detailed in the pseudocode below.

2. Additional Metrics Measured:

• CPU power consumption

• GPU power consumption

• Memory usage

Test Scenarios

1. Scenarios Tested:

• Power consumption without an ad-blocker (baseline scenario).

• Power consumption with different ad-blockers: AdBlock, AdBlock Plus, uBlock, uBlock Origin, and uBlock Origin Lite.

• Each scenario was tested on both Windows and Ubuntu systems.

2. Number of Repetitions for Each Scenario:

• The same video was played from the following websites for each ad-blocker, and each test was repeated three times for statistical significance:

– 9gag

– ARYzap

– Dailymotion

– Kisscartoon

– YouTube

– Cricbuzz

– Espncricinfo

– TheNews

3. Comparisons Performed:

• Ad Blockers on Different Websites: Power consumption data was collected for each ad-blocker across all the aforementioned websites. This allowed for a comparison of energy efficiency across different ad-blockers when exposed to varying content types.

• Windows vs. Ubuntu: The experiment also aimed to compare system power consumption between Windows and Ubuntu operating systems for each scenario, highlighting any differences in how the two systems handle advertisements and ad-blocking software.

Data Collection Procedure

1. Duration of Each Test Run: The duration of each test run depended on the length of the video, but the same video length was used for each ad-blocker scenario to ensure consistency.

2. Logging Method: The data collection procedure for Ubuntu involved a custom script to log CPU and GPU power consumption along with memory usage. The pseudocode in Fig. 1 details the methodology used for logging the data.

Fig. 1. Details of the methodology used for logging the data.

Results

The results of this study are presented to evaluate the impact of different ad-blockers on system power consumption for systems with an AI accelerator, comparing the performance on both Windows and Ubuntu operating systems. Specifically, comparisons were made between power consumption without an ad-blocker and with different ad-blockers (AdBlock, AdBlock Plus, uBlock, uBlock Origin, and uBlock Origin Lite) across various websites.

CPU Power Consumption by Ad-Blockers on Windows

Table I shows CPU power consumption (in watts) for different websites using various ad-blockers, including Adblock, Adblock Plus, uBlock, uBlock Origin, and uBlock Origin Lite. The goal is to compare the effectiveness of these ad-blockers in reducing CPU power usage on Windows.

Websitws Without adblock With adblock With adblockplus With ublock With ublock origin With ublock origin lite
9gag 11.2 9.1 10.4 8.5 8.2 8.2
ARYzap 10.3 2.5 10.0 10.0 7.5 9.5
Dailymotion 17.9 10.4 10.6 9.6 10.5 10.0
Kisscartoon 12.4 9.6 10.7 9.2 9.1 9.2
Youtube 4.0 3.8 4.1 5.7 4.2 4.1
Cricbuzz 11.7 11.1 9.6 10.4 9.8 9.6
Espncricinfo 18.6 18.6 17.5 17.4 17.6 17.5
TheNews 11.0 11.2 12.0 10.7 10.8 10.6
Table I. CPU Power Consumption by Ad-Blockers on Windows

uBlock Origin and uBlock Origin Lite consistently resulted in the lowest CPU power consumption across most websites, showing their effectiveness in reducing power usage. For example, 9gag showed a reduction from 11.2 W (without an ad-blocker) to 8.2 W with uBlock Origin. In contrast, Adblock Plus demonstrated moderate reductions, while Adblock had mixed performance, especially with ARYzap, where it reduced consumption to 2.5 W.

A stacked bar chart is proposed to illustrate the comparative effectiveness of different ad-blockers in reducing CPU power consumption across multiple websites.

This visualization highlights the overall reduction in power usage achieved by each ad-blocker and provides insights into which websites contribute most to the total power savings. The stacked bar chart effectively conveys both the aggregate impact and individual website-specific efficiency of each ad-blocker, making it a comprehensive tool for comparative analysis.

GPU Power Consumption by Ad-Blockers on Windows

This table presents GPU power consumption (in watts) across different websites using various ad-blockers, including Adblock, Adblock Plus, uBlock, uBlock Origin, and uBlock Origin Lite. The objective is to evaluate the effectiveness of these ad-blockers in reducing GPU power consumption on the Windows operating system.

From Table II, uBlock Origin demonstrated the most consistent reduction in GPU power consumption across most websites, such as 9gag and Dailymotion, with notable reductions from 25.3 W (without an ad-blocker) to 21.5 W and 24.1 W, respectively. Adblock and Adblock Plus generally showed moderate reductions, while uBlock had mixed results, including increased GPU consumption on Espncricinfo (41.3 W). YouTube showed a reduction in GPU power from 11.8 W without an ad-blocker to 9.9 W with Adblock, indicating its efficiency for lighter media content.

Websites Without adblock With adblock With adblockplus With ublock With ublock origin With ublock origin lite
9gag 25.3 25.0 24.6 23.8 21.5 23.0
Aryzap 26.4 24.2 25.7 22.2 22.2 25.1
Dailymotion 34.4 28.0 27.6 27.2 24.1 26.0
Kisscartoon 28.6 23.2 25.5 25.1 24.2 23.6
Youtube 11.8 9.9 12.8 14.3 13.0 13.5
Cricbuzz 29.2 28.3 25.1 27.8 24.2 24.2
Espncricinfo 25.0 29.0 29.0 41.3 36.4 37.7
Thenews 35.0 27.3 26.9 32.5 25.3 26.3
Table II. GPU Power Consumption by Ad-Blockers on Windows

Fig. 2 is proposed to illustrate the comparative effectiveness of different ad-blockers in reducing GPU power consumption across multiple websites.

Fig. 2. Comparison of CPU power consumption by ad-blockers on windows.

CPU Power Consumption by Ad-Blockers on Ubuntu

Table III presents CPU power consumption (in watts) across different websites when using various ad-blockers, including Adblock, Adblock Plus, uBlock, uBlock Origin, and uBlock Origin Lite, on the Ubuntu operating system. Fig. 4 compares the effectiveness of these ad-blockers in reducing CPU power usage on Ubuntu.

Websites Without adblock With adblock With adblockplus With ublock With ublock origin With ublock origin lite
9gag 20.36585 20.36585 13.60976 13.36585 12.175 10.15
ARYzap 10.83077 12.00833 12.10833 12.26667 12.39167 10.325
Dailymotion 21.84566 12.91489 14.19149 12.55239 11.38298 13
Kisscartoon 12.62069 16.43704 15.68302 15.87925 16.77358 16.21509
Youtube 12.62069 16.43704 15.68302 15.87925 16.77358 16.21509
Cricbuzz 19.55833 13.40833 15.725 13.925 14.68333 13.55833
Espncricinfo 23.53623 22.15942 22.62319 23.5942 21.85507 21.2029
TheNews 12.68333 14.7 13.40833 14.45833 13.44167 14.36667
Table III. CPU Power Consumption by Ad-Blockers on Ubuntu

Fig. 4. Comparison of CPU power consumption by ad-blockers on Ubuntu.

uBlock Origin Lite and uBlock Origin achieved the lowest CPU power consumption across several websites, notably on 9gag and Dailymotion, reducing the power from 20.37 W (without an ad-blocker) to 10.15 W and 11.38 W, respectively. In contrast, Adblock Plus and uBlock showed mixed performance, with some websites (e.g., Espncricinfo) experiencing slightly increased power consumption compared to no ad-blocker. This indicates that uBlock Origin Lite was the most energy-efficient option overall on Ubuntu.

Fig. 3 is used to illustrate the comparative effectiveness of different ad-blockers in reducing CPU power consumption across multiple websites.

Fig. 3. Comparision of GPU power consumption by ad-blockers on windows.

GPU Power Consumption by Ad-Blockers on Ubuntu

Table IV presents GPU power consumption (in watts) across different websites when using various ad-blockers, including Adblock, Adblock Plus, uBlock, uBlock Origin, and uBlock Origin Lite, on the Ubuntu operating system. Fig. 4 compares the effectiveness of these ad-blockers in reducing GPU power usage on Ubuntu.

Websites Without adblock With adblock With adblockplus With ublock With ublock origin With ublock origin lite
9gag 28.24697 15.54487 12.68321 13.96951 13.11368 11.95289
ARYzap 11.46285 17.0395 11.42833 9.560917 9.643083 8.68975
Dailymotion 19.6329 13.51326 8.478333 11.70014 11.39044 14.532
Kisscartoon 12.48525 11.95083 8.859167 11.3735 9.77025 9.95225
Youtube 15.2075 24.37908 23.88533 24.1685 24.80083 24.63283
Cricbuzz 28.932 15.12467 18.04992 13.96042 14.82758 13.6695
Espncricinfo 35.55667 32.83754 33.2113 35.75812 33.97362 33.5758
TheNews 12.3875 16.39342 16.85258 20.1735 16.59958 17.98608
Table IV. GPU Power Consumption by Ad-Blockers on Ubuntu

In Table IV, uBlock Origin Lite and Adblock Plus demonstrated the most consistent reduction in GPU power consumption for most websites, particularly for 9gag and Kisscartoon, with reductions from 28.25 W (without an ad-blocker) to 11.95 W and 8.86 W, respectively. ARYzap also showed the lowest GPU power consumption with uBlock Origin Lite at 8.69 W. Conversely, YouTube did not show any significant reduction, and GPU power consumption remained similar across all ad-blockers.

Fig. 5 provides a visual comparison of the power consumption data across all ad-blockers and websites.

Fig. 5. Comparison of GPU power consumption by ad-blockers on Ubuntu.

Memory Usage by Ad-Blockers on Windows (in MBs)

Table V presents the memory usage (in megabytes) for different websites while using various ad-blockers, including Adblock, Adblock Plus, uBlock, uBlock Origin, and uBlock Origin Lite, on the Windows operating system. The aim is to compare the effectiveness of these ad-blockers in reducing memory usage, which indicates the impact of each ad-blocker on system resource efficiency.

Websites Without adblock With adblock With adblockplus With ublock With ublock origin With ublock origin lite
9gag 8106.538 8171.436 8156.769 7886.61 7799.132 7786.368
ARYzap 10592.24 10522.94 10578.93 10497.67 10329.94 10299.97
Dailymotion 11378.77 9039.435 8975.851 8935.417 10452.69 10382.44
Kisscartoon 8014.356 8158.286 8099.026 7699.921 7759.095 7776.5
Youtube 5535.633 5796.438 5879.622 5821.586 10375.21 10342.7
Cricbuzz 8065.493 7958.654 7965.456 7862.528 7902.571 7957.448
Espncricinfo 8372.398 8372.398 8059 7892.362 7858.898 7693.081
TheNews 7975.206 8033.767 8139.765 7880.556 7793.475 7759.271
Table V. Memory Usage by Ad-Blockers on Windows (in MBs)

uBlock Origin and uBlock Origin Lite resulted in the most consistent reduction in memory usage across most websites. For instance, 9gag saw a reduction in memory usage from 8106.54 MB (without an ad-blocker) to 7786.37 MB with uBlock Origin Lite. Kisscartoon also saw a decrease, with uBlock resulting in the lowest memory usage at 7699.92 MB. Conversely, YouTube exhibited increased memory usage for most ad-blockers, with uBlock Origin reaching 10375.21 MB, highlighting potential inefficiencies for multimedia-rich content.

Bar chart has been used to visually compare memory usage for each ad-blocker across different websites (Fig. 6).

Fig. 6. Comparison of memory usage by ad-blockers on Windows [MBs].

Memory Usage by Ad-Blockers on Ubuntu (in MB)

This table presents the memory usage (in megabytes) for different websites while using various ad-blockers, including Adblock, Adblock Plus, uBlock, uBlock Origin, and uBlock Origin Lite, on the Ubuntu operating system. The goal is to compare the effectiveness of these ad-blockers in reducing memory usage, which provides insight into the system resource efficiency of each ad-blocker.

Table VI, uBlock Origin and uBlock Origin Lite demonstrated the lowest memory usage across multiple websites, particularly for 9gag and Kisscartoon, with reductions from 3248.28 MB (without an ad-blocker) to 2876.96 MB and 2960.17 MB, respectively. Dailymotion also showed significant reductions with uBlock at 3187.80 MB. In contrast, YouTube and Cricbuzz displayed relatively stable or increased memory usage across all ad-blockers, with no major savings observed.

Websites Without adblock With adblock With adblockplus With ublock With ublock origin With ublock origin lite
9gag 3248.28 3248.28 3330 3348.32 2876.96 2942.08
ARYzap 3347.077 3341 3516.033 3417.592 3333.267 3277.467
Dailymotion 3552.713 3532.893 3407.667 3187.8 3350.987 3472.947
Kisscartoon 3400.978 3106.45 3129.908 3086.767 3029.767 2960.167
Youtube 3353.459 3641.898 3551.536 3561.985 3561.691 3478.563
Cricbuzz 3611.075 3675.842 3745.183 3675.492 3656.85 3669.558
Espncricinfo 3313.812 3444.348 3864.536 3692.739 3612.638 3128.754
TheNews 3506.458 3658.633 3672.583 3698.392 3714.283 3534.683
Table VI. Memory Usage by Ad-Blockers on Ubuntu (in MB)

A stacked bar chart has been employed to illustrate the comparative memory usage of different ad-blockers across multiple websites (Fig. 7).

Fig. 7. Comparison of memory usage by ad-blockers on Ubuntu [MBs].

The results of this study show that uBlock Origin and uBlock Origin Lite are the most effective ad-blockers for reducing CPU and GPU power consumption as well as memory usage across both Windows and Ubuntu. On Windows, uBlock Origin and uBlock Origin Lite reduced CPU power consumption for 9gag by 26.8%, from 11.2 W to 8.2 W, and for Kisscartoon by 26.6%, from 12.4 W to 9.1 W. On Ubuntu, these ad-blockers performed even better, reducing CPU power consumption for 9gag by 50.2%, from 20.37 W to 10.15 W, and for Dailymotion by 47.9%, from 21.85 W to 11.38 W.

In terms of GPU power consumption, uBlock Origin and uBlock Origin Lite also achieved significant reductions. On Windows, Dailymotion’s GPU consumption decreased by 30%, from 34.4 W to 24.1 W. On Ubuntu, uBlock Origin Lite reduced 9gag’s GPU consumption by 57.7%, from 28.25 W to 11.95 W. However, for multimedia-rich websites like YouTube, GPU power consumption increased with most ad-blockers, indicating that blocking dynamic video ads can add processing overhead.

For memory usage, uBlock Origin and uBlock Origin Lite consistently reduced consumption. On Windows, Espncricinfos memory usage decreased by 8.1%, from 8372.40 MB to 7693.08 MB with uBlock Origin Lite. On Ubuntu, Kisscartoons memory usage dropped by 12.9%, from 3400.98 MB to 2960.17 MB with uBlock Origin Lite. However, for YouTube, memory usage increased significantly, indicating potential inefficiencies in handling complex ad content.

Overall, uBlock Origin and uBlock Origin Lite were the most efficient in reducing power and memory usage, with Ubuntu generally showing better results than Windows. Multimedia-heavy websites, however, presented challenges, often resulting in increased resource usage due to the processing overhead involved in blocking dynamic ads. These results highlight the importance of choosing the right ad-blocker based on the type of content and operating system to achieve the best energy efficiency and resource optimization.

Conclusion and Future Work

The findings from this study provide valuable insights into the effects of ad-blockers on system power consumption in computing environments equipped with AI accelerators. uBlock Origin and uBlock Origin Lite consistently outperformed other ad-blockers, resulting in the most substantial reductions in both CPU and GPU power consumption as well as memory usage on both Windows and Ubuntu. On Windows, the CPU power consumption for websites like 9gag was reduced by 26.8% using uBlock Origin. Similarly, GPU power consumption on Ubuntu saw a 57.7% reduction for 9gag using uBlock Origin Lite. These results demonstrate that AI accelerators, when combined with the right ad-blockers, can effectively optimize power usage. However, the analysis also highlighted that multimedia-rich websites, such as YouTube, did not experience significant power reductions. Instead, resource usage increased due to the overhead of processing dynamic video content. This suggests that ad-blockers introduce additional processing requirements when handling multimedia advertisements, which can lead to inefficiencies.

Overall, the study concludes that uBlock Origin and uBlock Origin Lite are the most effective ad-blockers in terms of energy efficiency. Moreover, Ubuntu generally demonstrated better results than Windows, indicating that operating system choice plays a significant role in resource optimization. The results emphasize the need for careful selection of ad-blockers based on content type and system architecture to achieve the best energy efficiency. This research has implications for developers of ad-blocking technologies, system manufacturers, and end-users seeking to maximize energy efficiency and system performance.

In terms of future work, while this study provided substantial insights into the impact of ad-blockers on system power consumption, there are several areas worth exploring further. Firstly, the impact of ad-blockers on a broader variety of hardware configurations, including systems with more advanced AI accelerators, should be investigated to generalize the findings further. Additionally, examining the effect of ad-blockers on other operating systems, such as macOS or mobile platforms, could provide a more comprehensive understanding of their effectiveness across different environments. Future studies could also focus on optimizing ad-blocking algorithms to minimize processing overhead, particularly for multimedia content, to ensure that energy-saving benefits extend to video-rich websites. Finally, a more detailed analysis of network traffic and data usage implications in combination with power metrics could provide a holistic view of the benefits and trade-offs of using ad-blockers in AI-accelerated systems.

References

  1. Mishra, Cha J, Park H, Kim S. Artificial Intelligence and Hardware Accelerators. Cham, Switzerland: Springer; 2023.
     Google Scholar
  2. Chen Y-H, Krishna T, Emer JS, Sze V. Eyeriss: an energy-efficient reconfigurable accelerator for deep convolutional neural networks. IEEE J Solid-State Circ. 2017 Jan;52(1):127–38.
     Google Scholar
  3. Choi S, Sim J, Kang M, Choi Y, Kim H, Kim LS. An energy-efficient deep convolutional neural network training accelerator for in situ personalization on smart devices. IEEE J Solid-State Circ. 2020 Oct;55(10):2691–702.
     Google Scholar
  4. Lee J, Kim C, Kang S, Lee S, Won J, Lee H, et al. UNPU: an energy-efficient deep neural network accelerator with fully variable weight bit precision. IEEE J Solid-State Circ. 2019 Jan;54(1):173–85.
     Google Scholar
  5. Song L, Qian X, Li H, Chen Y. Pipelayer: a pipelined ReRAM-based accelerator for deep learning. Proceedings of the IEEE International Symposium High Performance Computer Architecture (HPCA), pp. 541–52, 2017.
     Google Scholar
  6. Cherupally SK, Ramaswamy N, Juma J, Ye Y, Yadav AV, Govindan R, et al. In-memory computing for AI accelerators: challenges and solutions. Semicond Sci Technol. 2022;37(3):034001.
     Google Scholar
  7. Kwon H, Samajdar A, Krishna T. MAERI: enabling flexible dataflow mapping over DNN accelerators via reconfigurable inter-connects. Proceedings of the 23rd International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS), pp. 461–75, 2018.
     Google Scholar
  8. Reuther A, Michaleas P, Jones M, Kepner J, Samsi S, Gadepally V. Survey of machine learning accelerators. Proceedings of the IEEE High Performance Extreme Computing Conference (HPEC), pp. 1–12, 2020.
     Google Scholar
  9. Jiang, Li J, Liu F, Gao W, Wang L, Lan C, et al. A systematic study on benchmarking AI inference accelerators. CCF Trans High Perform Comput. 2022;4:87–103.
     Google Scholar
  10. Capra M, Bussolino B, Marchisio A, Shafique M, Masera G. An updated survey of efficient hardware architectures for accelerating deep convolutional neural networks. Future Internet. 2020 Jul;12(7):113.
     Google Scholar
  11. Capra M, Bussolino B, Marchisio A, Macii E, Martina M, Masera G, et al. Hardware and software optimizations for accelerating deep neural networks: survey of current trends, challenges, and the road ahead. IEEE Trans Comput. 2021 Aug;70(8):1197–212.
     Google Scholar
  12. Azghadi MR, Lammie C, Eshraghian JK, Wang L, Kavehei O. Hardware implementation of deep network accelerators towards healthcare and biomedical applications. IEEE Trans Circ Syst. 2020 Aug;67(8):2626–37.
     Google Scholar
  13. Himeur M, Bensaali B, Amira A, Bourguiba A, Elhoseny M, Karray F. AI-empowered methods for smart energy consumption: a review of load forecasting, anomaly detection, and demand response. Int J Precis Eng Manuf-Green Technol. 2023 Jan;10(1):25–45.
     Google Scholar
  14. Mirhoseini A, Goldie A, Yazgan M, Jiang JW, Songhori E, Wang S, et al. A graph placement methodology for fast chip design. Nature. 2021 Jun;594:207–12.
     Google Scholar
  15. Ke L, Gupta U, Cho BY, Brooks D, Wu C, Lee HS. RecNMP: accelerating personalized recommendation with near-memory processing. Proceedings pf the ACM/IEEE 47th Annual International Symposium on Computer Architecture (ISCA), pp. 790–803, 2020.
     Google Scholar
  16. Ray A, Ghasemkhani H, Kannan KN. Ad-blockers, advertisers, and internet: the economic implications of ad-blocker platforms. Proceedings of the International Conference on Information Systems (ICIS), 2017.
     Google Scholar
  17. Datta B, Madio L. Effects of ad-blockers adoption on digital piracy: a blessing or a curse? LCII Working Paper 2017-5, 2017.
     Google Scholar
  18. Frik A, Haviland A, Acquisti A. The impact of ad-blockers on consumer behavior: a lab experiment. Proceedings of the 18th Workshop on the Economics of Information Security (WEIS), 2019.
     Google Scholar
  19. Aseri M, Dawande M, Janakiraman G. Ad-blockers: a blessing or a curse? Inf Syst Res. 2020 Sep;31(3):889–908.
     Google Scholar
  20. Pearce JM. Energy conservation with open source ad-blockers. Technologies. 2020 Jun;8(2):27.
     Google Scholar
  21. Heitmann N, Pirker B, Park S. Towards building better mobile web browsers for ad blocking: the energy perspective. Proceedings of the ACM SIGPLAN International Symposium on New Ideas, New Paradigms, and Reflections on Programming and Software (Onward!), pp. 80–90, 2020.
     Google Scholar
  22. Albasir A, Naik K, Plourde B. Experimental study of energy and bandwidth costs of web advertisements on smartphones. Proceedings of the 11th IEEE International Conference on Mobile Ad Hoc and Sensor Systems (MASS), pp. 693–8, 2014.
     Google Scholar