Pub Date : 2019-10-01DOI: 10.1109/IGSC48788.2019.8957200
Zachary Gazzillo, Scott V. Franklin, S. L. Alarcón
We reformulate the problem of granular random close packing of 2D discs as a Quadratic Unconstrained Binary Optimization in order to utilize the D-Wave 2000Q quantum annealing computer. The solution is a set of ground states corresponding to jammed configurations in which no single particle can be moved without creating a non-zero potential. The problem is adapted to the quantum computer by discretizing space and mapping each point onto physical quantum-bits (qubits). An objective function is derived that defines the system energy for arbitrary particle locations, subject to constraints biasing solutions toward a pre-determined number of particles. Uniquely, the quantum computer samples and returns minimum values of this function finding low energy states, a subset of which are physically realizable solutions we seek. While quantum computing's technological infancy restricts our study to proofof-concept, our work still shows promise for efficient analysis of complex granular problems.
{"title":"Using Quantum Computers to Study Random Close Packing of Granular Discs","authors":"Zachary Gazzillo, Scott V. Franklin, S. L. Alarcón","doi":"10.1109/IGSC48788.2019.8957200","DOIUrl":"https://doi.org/10.1109/IGSC48788.2019.8957200","url":null,"abstract":"We reformulate the problem of granular random close packing of 2D discs as a Quadratic Unconstrained Binary Optimization in order to utilize the D-Wave 2000Q quantum annealing computer. The solution is a set of ground states corresponding to jammed configurations in which no single particle can be moved without creating a non-zero potential. The problem is adapted to the quantum computer by discretizing space and mapping each point onto physical quantum-bits (qubits). An objective function is derived that defines the system energy for arbitrary particle locations, subject to constraints biasing solutions toward a pre-determined number of particles. Uniquely, the quantum computer samples and returns minimum values of this function finding low energy states, a subset of which are physically realizable solutions we seek. While quantum computing's technological infancy restricts our study to proofof-concept, our work still shows promise for efficient analysis of complex granular problems.","PeriodicalId":399970,"journal":{"name":"2019 Tenth International Green and Sustainable Computing Conference (IGSC)","volume":"101 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127185022","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-10-01DOI: 10.1109/IGSC48788.2019.8957172
Mohit Kumar, Weisong Shi
In 2018 Turing Award lecture, John L. Hennessy discusses software-centric opportunities to save Moore’s law. The software is a major consumer of energy in ICT, IoT, and edge systems, but even then the research to make it energy efficient remains fractional. Java is one of the most commonly used languages to develop software for these systems. Java has various command-line options that an application user can use for JVM tuning to enhance the performance of an application. However, there is no study about how these Java command-line options impact the energy consumption of an application. In this work, we explore the impact of various Java command-line options on SPECjvm2008 benchmarks in terms of energy consumption and execution time using different JDKs. Our key findings are: 1) Oracle JDK is more energy efficient than Open JDK, 2) Xint command-line option is the least energy efficient, 3) UseG1GC command-line option is the most energy efficient, and 4) Active energy and execution time show a high correlation.
在2018年的图灵奖演讲中，John L. Hennessy讨论了以软件为中心的机会来拯救摩尔定律。在信息通信技术、物联网和边缘系统中，软件是能源的主要消耗者，但即便如此，使其节能的研究仍然很少。Java是为这些系统开发软件最常用的语言之一。Java有各种命令行选项，应用程序用户可以使用这些选项进行JVM调优，以增强应用程序的性能。但是，没有关于这些Java命令行选项如何影响应用程序能耗的研究。在本文中，我们将探讨各种Java命令行选项对SPECjvm2008基准测试在能耗和使用不同jdk的执行时间方面的影响。我们的主要发现是:1)Oracle JDK比Open JDK更节能，2)Xint命令行选项最节能，3)UseG1GC命令行选项最节能，4)Active energy和执行时间表现出高度相关。
{"title":"Energy Consumption Analysis of Java Command-line Options","authors":"Mohit Kumar, Weisong Shi","doi":"10.1109/IGSC48788.2019.8957172","DOIUrl":"https://doi.org/10.1109/IGSC48788.2019.8957172","url":null,"abstract":"In 2018 Turing Award lecture, John L. Hennessy discusses software-centric opportunities to save Moore’s law. The software is a major consumer of energy in ICT, IoT, and edge systems, but even then the research to make it energy efficient remains fractional. Java is one of the most commonly used languages to develop software for these systems. Java has various command-line options that an application user can use for JVM tuning to enhance the performance of an application. However, there is no study about how these Java command-line options impact the energy consumption of an application. In this work, we explore the impact of various Java command-line options on SPECjvm2008 benchmarks in terms of energy consumption and execution time using different JDKs. Our key findings are: 1) Oracle JDK is more energy efficient than Open JDK, 2) Xint command-line option is the least energy efficient, 3) UseG1GC command-line option is the most energy efficient, and 4) Active energy and execution time show a high correlation.","PeriodicalId":399970,"journal":{"name":"2019 Tenth International Green and Sustainable Computing Conference (IGSC)","volume":"16 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128583663","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-10-01DOI: 10.1109/IGSC48788.2019.8957166
A. Roy, Hakan Aydin, Dakai Zhu
Heterogeneous multicore systems have been recently received much attention due to their power efficiency and ability to handle different workloads. In this paper, we consider real-time tasks with precedence constraints and fault tolerance requirements, and investigate how they can be implemented on heterogeneous dual-core systems in energy-aware fashion. Our framework is able to tolerate one transient fault per task, and one permanent processing core fault simultaneously. We develop a number of task partitioning, ordering, and frequency assignment techniques for energy efficiency. Our experimental results indicate that the proposed techniques significantly reduce energy consumption while satisfying the fault tolerance requirements.
{"title":"Energy-Efficient Fault Tolerance for Real-Time Tasks with Precedence Constraints on Heterogeneous Multicore Systems","authors":"A. Roy, Hakan Aydin, Dakai Zhu","doi":"10.1109/IGSC48788.2019.8957166","DOIUrl":"https://doi.org/10.1109/IGSC48788.2019.8957166","url":null,"abstract":"Heterogeneous multicore systems have been recently received much attention due to their power efficiency and ability to handle different workloads. In this paper, we consider real-time tasks with precedence constraints and fault tolerance requirements, and investigate how they can be implemented on heterogeneous dual-core systems in energy-aware fashion. Our framework is able to tolerate one transient fault per task, and one permanent processing core fault simultaneously. We develop a number of task partitioning, ordering, and frequency assignment techniques for energy efficiency. Our experimental results indicate that the proposed techniques significantly reduce energy consumption while satisfying the fault tolerance requirements.","PeriodicalId":399970,"journal":{"name":"2019 Tenth International Green and Sustainable Computing Conference (IGSC)","volume":"63 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129198286","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-10-01DOI: 10.1109/IGSC48788.2019.8957181
M. Gadou, Sankeerth Reddy Mogili, Tania Banerjee-Mishra, S. Ranka
Energy efficiency and power minimization have become critical for high performance computing systems. Most modern processors and co-processors are equipped with dynamic voltage and frequency scaling (DVFS) mechanisms where the operating frequency of a processor can be changed to lower power consumption. Additionally, only a subset of processors can be used to save overall energy when the scaling of the application does match with the increase in power requirements. In this paper, we present a systematic approach for deriving energy performance trade-offs on a hybrid multicore (CPU+GPU) processor. Using a proxy application for compressible multiphase turbulence that is representative of a large number of applications, we show how to derive energy performance trade-offs on a server consisting of multicore and multiple GPU processors.
{"title":"Multi-objective Optimization on DVFS based Hybrid Systems","authors":"M. Gadou, Sankeerth Reddy Mogili, Tania Banerjee-Mishra, S. Ranka","doi":"10.1109/IGSC48788.2019.8957181","DOIUrl":"https://doi.org/10.1109/IGSC48788.2019.8957181","url":null,"abstract":"Energy efficiency and power minimization have become critical for high performance computing systems. Most modern processors and co-processors are equipped with dynamic voltage and frequency scaling (DVFS) mechanisms where the operating frequency of a processor can be changed to lower power consumption. Additionally, only a subset of processors can be used to save overall energy when the scaling of the application does match with the increase in power requirements. In this paper, we present a systematic approach for deriving energy performance trade-offs on a hybrid multicore (CPU+GPU) processor. Using a proxy application for compressible multiphase turbulence that is representative of a large number of applications, we show how to derive energy performance trade-offs on a server consisting of multicore and multiple GPU processors.","PeriodicalId":399970,"journal":{"name":"2019 Tenth International Green and Sustainable Computing Conference (IGSC)","volume":"70 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126326721","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-10-01DOI: 10.1109/IGSC48788.2019.8957195
Mingming Guo, N. Pissinou, S. S. Iyengar
Big data analytics are pervasive in data-intensive systems and applications using machine learning and deep learning. IoT sensor devices are generating all types of big data, including structured (e.g., tables) and unstructured data (e.g., text and image), which are beyond the processing power of humans. However, how to conduct big data analysis on IoT data without compromising IoT privacy is still an open problem. IoT shall leverage powerful learning techniques to automatically learn patterns such as similarities, correlations and abnormalities from big sensing data in a privacy-preserving manner. To make this happen, we first examine distributed learning techniques that are suitable for IoT architectures. We then propose a privacy-preserving distributed learning framework with a novel dynamic deep learning mechanism to extract patterns and learn knowledge from IoT data. Simulations are performed to show the effectiveness and efficiency of our solution.
{"title":"Privacy-Preserving Deep Learning for Enabling Big Edge Data Analytics in Internet of Things","authors":"Mingming Guo, N. Pissinou, S. S. Iyengar","doi":"10.1109/IGSC48788.2019.8957195","DOIUrl":"https://doi.org/10.1109/IGSC48788.2019.8957195","url":null,"abstract":"Big data analytics are pervasive in data-intensive systems and applications using machine learning and deep learning. IoT sensor devices are generating all types of big data, including structured (e.g., tables) and unstructured data (e.g., text and image), which are beyond the processing power of humans. However, how to conduct big data analysis on IoT data without compromising IoT privacy is still an open problem. IoT shall leverage powerful learning techniques to automatically learn patterns such as similarities, correlations and abnormalities from big sensing data in a privacy-preserving manner. To make this happen, we first examine distributed learning techniques that are suitable for IoT architectures. We then propose a privacy-preserving distributed learning framework with a novel dynamic deep learning mechanism to extract patterns and learn knowledge from IoT data. Simulations are performed to show the effectiveness and efficiency of our solution.","PeriodicalId":399970,"journal":{"name":"2019 Tenth International Green and Sustainable Computing Conference (IGSC)","volume":"48 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126521060","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-10-01DOI: 10.1109/igsc48788.2019.8957176
{"title":"IGSC 2019 Copyright Page","authors":"","doi":"10.1109/igsc48788.2019.8957176","DOIUrl":"https://doi.org/10.1109/igsc48788.2019.8957176","url":null,"abstract":"","PeriodicalId":399970,"journal":{"name":"2019 Tenth International Green and Sustainable Computing Conference (IGSC)","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128064255","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-10-01DOI: 10.1109/IGSC48788.2019.8957183
Jacob M. Hope, Trisha Nag, Apan Qasem
This paper presents a new approach to extracting improved performance-per-watt on large-scale hybrid graph applications with sparse data access patterns. The proposed technique takes advantage of demand paging, a technology recently introduced on CPU-GPU systems with heterogeneous memory. The strategy combines an analytical cost model, compiler transformations and a runtime system. The cost model, guided by runtime feedback, judiciously selects data structures for host placement which are migrated to the GPU during kernel execution via demand paging. We then introduce, two new code transformations, kernel blocking and compute colocation, to exploit page-level locality in host-resident data.We evaluate our strategy on four important algorithms in graph analytics: BFS, MST, SSSP and PageRank. Demand paging combined with kernel blocking causes significant reduction in PCIe traffic and yields an average speedup of 2.46, and up to a 5 $times $ performance improvement on BFS, over state-of-the-art methods. The performance boost does not incur a commensurate increase in GPU power draw, thereby leading to significant gains in energy efficiency. On average, 2.36 improvement in performance-per-watt is achieved across the four algorithms.
{"title":"Energy-Efficient GPU Graph Processing with On-Demand Page Migration","authors":"Jacob M. Hope, Trisha Nag, Apan Qasem","doi":"10.1109/IGSC48788.2019.8957183","DOIUrl":"https://doi.org/10.1109/IGSC48788.2019.8957183","url":null,"abstract":"This paper presents a new approach to extracting improved performance-per-watt on large-scale hybrid graph applications with sparse data access patterns. The proposed technique takes advantage of demand paging, a technology recently introduced on CPU-GPU systems with heterogeneous memory. The strategy combines an analytical cost model, compiler transformations and a runtime system. The cost model, guided by runtime feedback, judiciously selects data structures for host placement which are migrated to the GPU during kernel execution via demand paging. We then introduce, two new code transformations, kernel blocking and compute colocation, to exploit page-level locality in host-resident data.We evaluate our strategy on four important algorithms in graph analytics: BFS, MST, SSSP and PageRank. Demand paging combined with kernel blocking causes significant reduction in PCIe traffic and yields an average speedup of 2.46, and up to a 5 $times $ performance improvement on BFS, over state-of-the-art methods. The performance boost does not incur a commensurate increase in GPU power draw, thereby leading to significant gains in energy efficiency. On average, 2.36 improvement in performance-per-watt is achieved across the four algorithms.","PeriodicalId":399970,"journal":{"name":"2019 Tenth International Green and Sustainable Computing Conference (IGSC)","volume":"2016 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131417435","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-10-01DOI: 10.1109/IGSC48788.2019.8957199
M. Haddad, J. Nicod, C. Varnier, Marie-Cécile Peéra
The trend toward server-side computing and the exploding popularity of Internet services due to the increasing of demand for networking, storage and computation has created a world-wild energetic problem and a significant carbon footprint. These environmental concerns prompt to several green energy initiative aiming either to increase data center efficiency and/or to the use of green energy supply. In this regard, As part of the ANR DATAZERO project, many researchers are working to define main concepts of an autonomous green data center only powered by renewable energies. Thus, the present paper proposes a mixed integer linear program to optimize the commitment of a hybrid energy system composed of wind turbines, solar panels, batteries and hydrogen storage systems. The approach is used to supply a data center demand and takes the weather forecasts into account at the time of optimization. Different time window resolution are applied in order to verify the best time window for decision making.
{"title":"Mixed Integer Linear Programming Approach to Optimize the Hybrid Renewable Energy System Management for supplying a Stand-Alone Data Center","authors":"M. Haddad, J. Nicod, C. Varnier, Marie-Cécile Peéra","doi":"10.1109/IGSC48788.2019.8957199","DOIUrl":"https://doi.org/10.1109/IGSC48788.2019.8957199","url":null,"abstract":"The trend toward server-side computing and the exploding popularity of Internet services due to the increasing of demand for networking, storage and computation has created a world-wild energetic problem and a significant carbon footprint. These environmental concerns prompt to several green energy initiative aiming either to increase data center efficiency and/or to the use of green energy supply. In this regard, As part of the ANR DATAZERO project, many researchers are working to define main concepts of an autonomous green data center only powered by renewable energies. Thus, the present paper proposes a mixed integer linear program to optimize the commitment of a hybrid energy system composed of wind turbines, solar panels, batteries and hydrogen storage systems. The approach is used to supply a data center demand and takes the weather forecasts into account at the time of optimization. Different time window resolution are applied in order to verify the best time window for decision making.","PeriodicalId":399970,"journal":{"name":"2019 Tenth International Green and Sustainable Computing Conference (IGSC)","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114368794","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-10-01DOI: 10.1109/IGSC48788.2019.8957165
Yi Zhou, Yuanqi Chen, Chaowei Zhang, X. Qin, Jifu Zhang
The energy efficiency of a data center depends on the cooling cost of clusters in the data center. Enhancing thermal efficiency of clusters is a practical approach to reducing energy consumption cost, optimizing scalability, and improving reliability. In this paper, we propose ThermoBench to evaluate the thermal efficiency of computing and storage clusters deployed in data centers. We shed light on the criteria and challenges of developing a thermal efficiency benchmark. We pay particular attention on clusters running scalable client-server enterprise applications in data centers. We characterize workload conditions in such a cluster computing environment in forms of client sessions of multiple requests. To resemble real-world applications, ThermoBench makes use of the TPCW benchmark to changes transaction mix and load over time. We apply ThermoBench to evaluate the thermal efficiency of a real-world cluster. Experimental results show that ThermalBench provides a simple yet powerful benchmark solution for assessing thermal behaviors of computing clusters in data centers.
{"title":"Thermal-Efficiency Benchmark on High-Performance Clusters","authors":"Yi Zhou, Yuanqi Chen, Chaowei Zhang, X. Qin, Jifu Zhang","doi":"10.1109/IGSC48788.2019.8957165","DOIUrl":"https://doi.org/10.1109/IGSC48788.2019.8957165","url":null,"abstract":"The energy efficiency of a data center depends on the cooling cost of clusters in the data center. Enhancing thermal efficiency of clusters is a practical approach to reducing energy consumption cost, optimizing scalability, and improving reliability. In this paper, we propose ThermoBench to evaluate the thermal efficiency of computing and storage clusters deployed in data centers. We shed light on the criteria and challenges of developing a thermal efficiency benchmark. We pay particular attention on clusters running scalable client-server enterprise applications in data centers. We characterize workload conditions in such a cluster computing environment in forms of client sessions of multiple requests. To resemble real-world applications, ThermoBench makes use of the TPCW benchmark to changes transaction mix and load over time. We apply ThermoBench to evaluate the thermal efficiency of a real-world cluster. Experimental results show that ThermalBench provides a simple yet powerful benchmark solution for assessing thermal behaviors of computing clusters in data centers.","PeriodicalId":399970,"journal":{"name":"2019 Tenth International Green and Sustainable Computing Conference (IGSC)","volume":"30 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132492965","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-10-01DOI: 10.1109/IGSC48788.2019.8957173
Corridon McKelvey, Richard Dreyer, Donnel Zhu, Wei Wang, J. Quarles
The advent of wearable devices introduces many opportunities with unconventional computing paradigms. In this work, we investigate energy-efficient designs of an augmented- reality application, FaceReminder, on a VUZIX Blade® smart glass by exploiting its unique see-through display and on-board camera. Powered with well-known facial recognition techniques, FaceReminder aims at helping people with prosopagnosia (i.e., face blindness) or short-memory of face-name connection by showing the names of the person on the see-through display. To cope with the limited resources (especially battery) of the smart glass, we explore designs that offload portion of the computation in facial detection and recognition to another mobile device (e.g., smart phone), which pairs with the glass via Bluetooth. Several optimization techniques, such as resolution adjustment and cropping, have been investigated to improve the latency and energy efficiency with reduced image sizes. We implemented FaceReminder and empirically evaluated its accuracy, latency and energy consumption of the major steps (including photo taking, resizing/cropping, Bluetooth transmission, facial detection and recognition). Compared to the baseline Glass-Only design, the most efficient Paired Glass-Device design with photos of reduced resolution and MTCNN facial detection technique can reduce the average latency by 73% and energy consumption by 78.9% (i.e., about 5X battery life improvement) while maintaining more than satisfactory 80% recognition accuracy.
{"title":"Energy-Oriented Designs of an Augmented-Reality Application on a VUZIX Blade Smart Glass","authors":"Corridon McKelvey, Richard Dreyer, Donnel Zhu, Wei Wang, J. Quarles","doi":"10.1109/IGSC48788.2019.8957173","DOIUrl":"https://doi.org/10.1109/IGSC48788.2019.8957173","url":null,"abstract":"The advent of wearable devices introduces many opportunities with unconventional computing paradigms. In this work, we investigate energy-efficient designs of an augmented- reality application, FaceReminder, on a VUZIX Blade® smart glass by exploiting its unique see-through display and on-board camera. Powered with well-known facial recognition techniques, FaceReminder aims at helping people with prosopagnosia (i.e., face blindness) or short-memory of face-name connection by showing the names of the person on the see-through display. To cope with the limited resources (especially battery) of the smart glass, we explore designs that offload portion of the computation in facial detection and recognition to another mobile device (e.g., smart phone), which pairs with the glass via Bluetooth. Several optimization techniques, such as resolution adjustment and cropping, have been investigated to improve the latency and energy efficiency with reduced image sizes. We implemented FaceReminder and empirically evaluated its accuracy, latency and energy consumption of the major steps (including photo taking, resizing/cropping, Bluetooth transmission, facial detection and recognition). Compared to the baseline Glass-Only design, the most efficient Paired Glass-Device design with photos of reduced resolution and MTCNN facial detection technique can reduce the average latency by 73% and energy consumption by 78.9% (i.e., about 5X battery life improvement) while maintaining more than satisfactory 80% recognition accuracy.","PeriodicalId":399970,"journal":{"name":"2019 Tenth International Green and Sustainable Computing Conference (IGSC)","volume":"51 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130842341","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：