The proceedings contain 5 papers. The topics discussed include: towards adaptive multi-alternative process network;BifurKTM: approximately consistent distributed transactional memory for GPUs;the impact of precision t...
ISBN:
(纸本)9783959771818
The proceedings contain 5 papers. The topics discussed include: towards adaptive multi-alternative process network;BifurKTM: approximately consistent distributed transactional memory for GPUs;the impact of precision tuning on embedded systems performance: a case study on field-oriented control;resource aware GPU scheduling in kubernetes infrastructure;and HPC application cloudification: the streamflow toolkit.
The optimised droop control method is proposed to achieve the state-of-charge (SoC) balance among parallel-connected distributed energy storage units in islanded DC microgrid, which considers the difference of line im...
详细信息
The optimised droop control method is proposed to achieve the state-of-charge (SoC) balance among parallel-connected distributed energy storage units in islanded DC microgrid, which considers the difference of line impedance, initial state-of-charge values and capacities among distributed energy storage units. Since the droop control is the basic control strategy for load sharing in DC microgrid applications, however, the load sharing accuracy is degraded under conventional droop control method due to the unmatched line impedance in reality. Meanwhile, the initial state-of-charge values and capacities of each distributed energy storage unit are usually different. Hence, the state of charge for distributed energy storage units cannot be balanced. In order to prolong the lifetime of the distributed energy storage units and avoid the overuse of a certain distributed energy storage unit, the optimised droop control strategy based on sample and holder is designed, by modifying the droop coefficient adaptively, the accurate load sharing and balanced state of charge among distributed energy storage units are both obtained. Finally, the performance of the proposed control scheme is accessed through a series cases on technologies real-time digital simulator (RTDS) and its effectiveness is verified.
This paper proposes an optimized framework for data lakes meant to enhance parameters like data rate, response time, and capacity in big data environments. Some common issues with traditional data lakes include issues...
详细信息
ISBN:
(数字)9798350388916
ISBN:
(纸本)9798350388923
This paper proposes an optimized framework for data lakes meant to enhance parameters like data rate, response time, and capacity in big data environments. Some common issues with traditional data lakes include issues of slow data access, and basic problem of scalability that becomes an issue as data sizes increase. These challenges are however countered by the proposed method through the use of distributed data structures of data, indexing quality, concurrent processing and in memory computing. The results obtained are 40% time saving in retrieving large data set and 50% improvement in query response time for heavy data. This in-memory processed system exhibited up to 60% gain in data throughput and achieved better scalability with shorter query response time as more parallel processing nodes were added. The evidences derived from the experiment clearly explain that the proposed method not only improves the performance but also offers cost efficient solution for the organizations those are dealing with big realtime data analytics. Subsequent studies may involve the exploration of higher levels of machine learning to enhance the predictability of the analytical approaches used and optimal resource allocation to improved data lake underway in the cloud.
Gaussian geostatistical space-time modeling is an effective tool for performing statistical inference of field data evolving in space and time, generalizing spatial modeling alone at the cost of the greater complexity...
详细信息
Recent developments in deep learning have significantly improved the quality of synthesized singing voice audio. However, prominent neural singing voice synthesis systems suffer from slow inference speed due to their ...
详细信息
ISBN:
(纸本)9781728163383
Recent developments in deep learning have significantly improved the quality of synthesized singing voice audio. However, prominent neural singing voice synthesis systems suffer from slow inference speed due to their autoregressive design. Inspired by MLP-Mixer, a novel architecture introduced in the vision literature for attention-free image classification, we propose MLP Singer, a parallel Korean singing voice synthesis system. To the best of our knowledge, this is the first work that uses an entirely MLP-based architecture for voice synthesis. Listening tests demonstrate that MLP Singer outperforms a larger autoregressive GAN-based system, both in terms of audio quality and synthesis speed. In particular, MLP Singer achieves a real-time factor of up to 200 and 3400 on CPUs and GPUs respectively, enabling order of magnitude faster generation on both environments.(1)
Digital image processing is widely used in various fields of science, such as medicine – X-ray analysis, magnetic resonance imaging, computed tomography, cosmology – collecting information from satellites, their tra...
详细信息
We present distributed task fusion, a run-time optimization for task-based runtimes operating on parallel and heterogeneous systems. distributed task fusion dynamically performs an efficient buffering, analysis, and f...
详细信息
ISBN:
(纸本)9781665454117
We present distributed task fusion, a run-time optimization for task-based runtimes operating on parallel and heterogeneous systems. distributed task fusion dynamically performs an efficient buffering, analysis, and fusion of asynchronously-evaluated distributed operations, reducing the overheads inherent to scheduling distributed tasks in implicitly parallel frameworks and runtimes. We identify the constraints under which distributed task fusion is permissible and describe an implementation in Legate, a domain-agnostic library for constructing portable and scalable task-based libraries. We present performance results using cuNumeric, a Legate library that enables scalable execution of NumPy pipelines on parallel and heterogeneous systems. We realize speedups up to 1.5x with task fusion enabled on up to 32 P100 GPUs, thus demonstrating efficient execution of pipelines involving many successive fine-grained tasks. Finally, we discuss potential future work, including complementary optimizations that could result in additional performance improvements.
The proceedings contain 16 papers. The topics discussed include: DPD-InfoGAN: differentially private distributed InfoGAN;towards optimal configuration of microservices;DistIR: an intermediate representation for optimi...
ISBN:
(纸本)9781450382984
The proceedings contain 16 papers. The topics discussed include: DPD-InfoGAN: differentially private distributed InfoGAN;towards optimal configuration of microservices;DistIR: an intermediate representation for optimizing distributed neural networks;towards a general framework for ML-based self-tuning databases;predicting CPU usage for proactive autoscaling;are we there yet? estimating training time for recommendation systems;Queen Jane approximately: enabling efficient neural network inference with context-adaptivity;AutoAblation: automated parallel ablation studies for deep learning;Vate: runtime adaptable probabilistic programming for Java;DISC: a dynamic shape compiler for machine learning workloads;and towards mitigating device heterogeneity in federated learning via adaptive model quantization.
Future Internet of Things (IoT)-driven applications will move from the cloud-centric IoT model to the hybrid distributed processing model, known as Fog computing, where some of the involved computational tasks (e.g. r...
详细信息
Future Internet of Things (IoT)-driven applications will move from the cloud-centric IoT model to the hybrid distributed processing model, known as Fog computing, where some of the involved computational tasks (e.g. real-time data analytics) are partially moved to the edge of the network to reduce latency and improve the network efficiency. In recent times, Fog computing has generated significant research interest for IoT applications, however, there is still a lack of ideal approach and framework for supporting parallel and fault-tolerant execution of the tasks while collectively utilizing the resource-constrained Fog devices. To address this issue, in this paper, we propose an Akka framework based on the Actor Model for designing and executing the distributed Fog applications. The Actor Model was conceived as a universal paradigm for concurrent computation with additional requirements such as resiliency and scalability, whereas, the Akka toolkit is a reference implementation of the model. Further, to dynamically deploy the distributed applications on the Fog networks, a Docker containerization approach is used. To validate the proposed actor-based framework, a wireless sensor network case study is designed and implemented for demonstrating the feasibility of conceiving applications on the Fog networks. Besides that, a detailed analysis is produced for showing the performance and parallelization efficiency of the proposed model on the resource-constrained gateway and Fog devices. (C) 2020 Elsevier B.V. All rights reserved.
暂无评论