To make parallel programming as widespread as parallel architectures, more structured parallel programming paradigms are necessary. One of the possible approaches are algorithmic skeletons. They can be seen as higher ...
详细信息
We consider the problem of scheduling an application composed of independent tasks on a fully heterogeneous master-worker platform with communication costs. We introduce a bi-criteria approach aiming at maximizing the...
详细信息
ISBN:
(纸本)9783642038686
We consider the problem of scheduling an application composed of independent tasks on a fully heterogeneous master-worker platform with communication costs. We introduce a bi-criteria approach aiming at maximizing the throughput of the application while minimizing the energy consumed by participating resources. Assuming arbitrary super-linear power consumption laws, we investigate different models for energy consumption, with and without start-tip overheads. Building upon closed-form expressions for the uniprocessor case;we derive optimal or asymptotically optimal solutions for both models.
Daryl Pregibon - Google Inc. Research Scientist states that: "Data Mining is a mixture of statistics, artificial intelligence and database research." In other words, the purpose of this process is the automa...
详细信息
ISBN:
(纸本)9781467361149
Daryl Pregibon - Google Inc. Research Scientist states that: "Data Mining is a mixture of statistics, artificial intelligence and database research." In other words, the purpose of this process is the automatic discovery of knowledge hidden in data using various computational techniques. The purpose of this work is represented by the analysis of the impact of GRID technology for storing and processing large amounts of information and knowledge. Using computational power of computers and the most effective means of working with data, information exploitation is no longer a difficulty. It shows a strong expansion of the use of GRID technologies in various fields, as a consequence of the development of our society and, in particular, of the scientific and technical world that require technologies that allow all parties to use resources in a well-controlled and well organized way. Therefore, we can use GRID technologies for Data Mining processing. To see what the data "mining" process consist of, we must go through the following steps: construction and validation of the model and application of the model to new data. GRID - Data Mining connection can be successfully used to monitor environmental factors in environmental protection field, in civil engineering field to monitor the behavior in time, in medical field to determine diagnoses, in telecommunications. To be able to develop "mining" applications of the distributed data within a GRID network, the infrastructure that will be used is the Knowledge GRID one. This high level infrastructure has an architecture dedicated to data "mining" operations and specialized services for resource discovery stored in distributed deposits, information services management. In this concept, the achievement of data storage and processing is one of the most effective ways one can obtain results with high accuracy, according to initial requirements, using the automated knowledge discovery principles from the entire resource of knowledge
Recent improvements in word representations (word embeddings) have improved a wide range of text-based information retrieval applications. Successfully representing many semantic characteristics of words in low dimens...
详细信息
ISBN:
(纸本)9781728129334;9781728129327
Recent improvements in word representations (word embeddings) have improved a wide range of text-based information retrieval applications. Successfully representing many semantic characteristics of words in low dimensional vector spaces with Continuous Bag of Words (CBOW) and Skip-Gram models invoked new techniques for textual input representations. In this study we introduce a neural embedding model for representing social media users using document representation model (doc2vec). We propose a simple method for evaluating the quality of the user vectors. We also share our results on simply averaging user vectors of the same category as category vectors. The experiment results show that our user2vec model creates semantically meaningful representations of users and it is very open for new improvements. We also share the dataset used in this study.
We present BSGP. a new programming language for general purpose computation on the GPU. A BSGP program looks much the same as a sequential C program. Programmers only need to supply a bare Minimum of extra information...
详细信息
We present BSGP. a new programming language for general purpose computation on the GPU. A BSGP program looks much the same as a sequential C program. Programmers only need to supply a bare Minimum of extra information to describe parallelprocessing on GPUs. As a result, BSGP programs are easy to read, write. and maintain. Moreover, the ease of programming does not come at the cost of performance. A well-designed BSGP compiler converts BSGP programs to kernels and combines them using optimally allocated temporary streams. In our benchmark, BSGP programs achieve similar or better performance than well-optimized CUDA programs. while the source code complexity and programming time are significantly reduced. To test BSGP's code efficiency and ease of programming, we implemented a variety of GPU applications, including a highly sophisticated X3D parser that would be extremely difficult to develop with existing GPU programming languages.
The Transmission Line Matrix Method (TLM) has been used extensively for modelling closed electromagnetic structures as by Johns. Johns Matrix techniques are now extending its applications to partially open structures,...
详细信息
The Transmission Line Matrix Method (TLM) has been used extensively for modelling closed electromagnetic structures as by Johns. Johns Matrix techniques are now extending its applications to partially open structures, and the modelling of planar antenna configurations becomes feasible. The TLM process yields the time response of a structure to impulsive excitation, which must usually be transformed into the frequency domain using Fourier transform techniques. Unfortunately, this requires a considerable number of computational steps, and then a considerable computational time, to yield satisfactory accuracy. In this paper we propose and describe two complementary procedures designed to significantly reduce computational expenditure. The first exploits the localised nature of the TLM process through parallelprocessing (implementation on the Connection Machine), and the other is based on the numerical processing of the TLM output using the Prony-Pisarenko Method. Both measures may be applied as well to accelerate other time domain methods such as FD-TD.
The authors describe a new approach to hardware implementation of speech compression algorithms. The hardware architecture takes into consideration future integration into the public telephone system or other multiuse...
详细信息
The authors describe a new approach to hardware implementation of speech compression algorithms. The hardware architecture takes into consideration future integration into the public telephone system or other multiuser systems. The authors propose a multichannel multiprocessor system placed at the central office rather than the traditional single-channel DSP-chip system. Dedicated processors and distributed and parallelprocessing are basic elements in the system architecture. The approach has been used to develop an experimental real-time processor for 16-kb/s compression, implementing the adaptive subbands excited transform (ASET) coding algorithm. A system capable of serving up to 24 channels has been successfully built and demonstrated. In addition to high-quality speech, the experimental system results in a cost per channel end that is approximately one-third the cost of a DSP-chip implementation.
The proceedings contain 53 papers. The topics discussed include: context-aware root cause localization in distributed traces using social network analysis;enhancing the performance of deep learning model based object ...
ISBN:
(纸本)9798400704451
The proceedings contain 53 papers. The topics discussed include: context-aware root cause localization in distributed traces using social network analysis;enhancing the performance of deep learning model based object detection using parallelprocessing;fair sharing of data in autotuning research;mastering computer vision inference frameworks;matrix network analyzer: a new decomposition algorithm for phase-type queueing networks;towards efficient diagnosis of performance bottlenecks in microservice-based applications;KubePlaybook: a repository of ansible playbooks for Kubernetes auto-remediation with LLMs;efficient unsupervised latency culprit ranking in distributed traces with GNN and critical path analysis;and network analysis of microservices: a case study on Alibaba production clusters.
Time critical applications are appealing to deploy in clouds due to the elasticity of cloud resources and their on-demand nature. However, support for deploying application components with strict deadlines on their de...
详细信息
ISBN:
(纸本)9783319642031;9783319642024
Time critical applications are appealing to deploy in clouds due to the elasticity of cloud resources and their on-demand nature. However, support for deploying application components with strict deadlines on their deployment is lacking in current cloud providers. This is particularly important for adaptive applications that must automatically and seamlessly scale, migrate, or recover swiftly from failures. A common deployment procedure is to transmit application packages from the application provider to the cloud, and install the application there. Thus, users need to manually deploy their applications into clouds step by step with no guarantee regarding deadlines. In this work, we propose a Deadline-aware Deployment System (DDS) for time critical applications in clouds. DDS enables users to automatically deploy applications into clouds. We design bandwidth-aware EDF scheduling algorithms in DDS that minimize the number of deployments that miss their deadlines and maximize the utilization of network bandwidth. In the evaluation, we show that DDS leverages network bandwidth sufficiently, and significantly reduces the number of missed deadlines during deployment.
The join operation is the most costly operation in relational database management systems. distributed and parallelprocessing can effectively speed up the join operation. In this paper, we describe a number of highly...
详细信息
暂无评论