There are increasing opportunities to calculate with distributed/parallel computing: many-core CPU, GPU, and FPGA. It is, however, generally difficult to come up with an algorithm suitable for distributed/parallel com...
详细信息
ISBN:
(纸本)9781728103921
There are increasing opportunities to calculate with distributed/parallel computing: many-core CPU, GPU, and FPGA. It is, however, generally difficult to come up with an algorithm suitable for distributed/parallel computing. There have been many researches on the automatic partitioning of programs and designs, but they have not performed reconstruction of the data flow in the algorithm level. In this paper, we apply partial synthesis method and synthesize algorithms for distributed/parallel environment automatically. We propose the template-based computing that aims to prevent the communications among the cores and chips to become overhead. With that template, an algorithm is synthesized by using conventional partial synthesis method, which performs synthesis iteratively modifying the template. The synthesis-problem generally becomes infeasible as its size becomes larger. Therefore, we propose the method to reduce the search space by adding constraints. The synthesis is performed for small instances of the target problem at first, and then the additional constraints are considered based on the synthesized algorithm. In the experiment, we synthesized algorithms for matrix vector multiplication with one-way ring -connected nodes to matrix of 32X32.
Tensor-based big data analysis approaches are effectively exploited to handle multisource and heterogeneous cyber-physical-social big data generated from diverse spaces. However, the curse of dimensionality seriously ...
详细信息
Tensor-based big data analysis approaches are effectively exploited to handle multisource and heterogeneous cyber-physical-social big data generated from diverse spaces. However, the curse of dimensionality seriously restricts their widespread exploitation, especially under edge/fog computing environments. To alleviate the dilemma, we attempt to present a set of tensor-train (TT)-based tensor operations with their scalable computations and then propose a novel TT-based big data processing framework under edge/fog computing environments. Specifically, in this article, we first summarize and present a set of TT-based tensor operations by converting the original high-order tensor operation to a series of low-order (second- or third-order) TTcore-based operations. Then, we propose a two-layer scalable TT-based computation architecture, including inter-TTcore and intra-TTcore scalable models. Afterward, according to various scalable models, a series of scalable TT-based tensor computations (STT-TCs) with their complexity analysis are proposed in detail. Finally, we propose a novel TT-based big data processing framework to adapt to edge/fog computing environments. We conduct extensive experiments based on both random data sets and real-world ubiquitous bus traffic data sets. Experimental results demonstrate that the proposed STT-TCs can significantly improve computation efficiency and are suitable for edge/fog computing environments.
With the advent of high-speed networks and the availability of powerful high-performance workstations, network of workstations has emerged as the most cost-effective platform for computation-intensive applications. On...
详细信息
With the advent of high-speed networks and the availability of powerful high-performance workstations, network of workstations has emerged as the most cost-effective platform for computation-intensive applications. One of the major applications for the network of workstations is in the field of remote sensing, where because of the high dimensionality of data, most of the existing data exploitation procedures are computation-intensive. To test the utility of the network of workstations in the field of remote sensing we have adopted a modified version of the well-known ISODATA classification procedure which may be considered as the benchmark for all unsupervised classification algorithms. The ISODATA algorithm is an iterative method that uses Enclidean distance as the similarity measure to cluster data elements into different classes. We have designed and developed a distributed version of ISODATA algorithm (D-ISODATA) on the network of workstations under a message-passing interface environment and have obtained promising speedup. To reduce the processing load and thereby increase the throughput, the ISODATA procedure is commonly applied to only the first few principal component images derived from the original set of the multispectral images. The drawback with the principal component approach is that it is based entirely on the statistical significance of the spectra, rather than the uniqueness of the individual spectra. Asl small objects and ground features would likely manifest themselves in the last principal component images, that is, eigen images, discarding them prior to classification would lead to the loss of valuable information. The significant enhancement in processing speed on the network of workstations makes it possible for us to apply our distributed algorithm D-ISODATA to the entire set of multispectral images directly, thereby preserving all the spectral signatures in the data, regardless of their statistical significance. (C) 1999 Academic Press.
Alternating Direction Implicit (ADI) methods have been in use since 1954 for the solution of both parabolic and elliptic partial differential equations. The convergence of these methods can be dramatically accelerated...
详细信息
ISBN:
(纸本)9780769550602
Alternating Direction Implicit (ADI) methods have been in use since 1954 for the solution of both parabolic and elliptic partial differential equations. The convergence of these methods can be dramatically accelerated when good estimates of the eigenvalues of the operator are available, However, in the case of computation on parallel computers, the solution of tridiagonal systems imposes an unreasonable overhead. We discuss methods to lower the overhead imposed by the solution of the corresponding tridiagonal systems. The proposed method has the same convergence properties as a standard ADI method, but all of the solves run in approximately the same time as the "fast" direction. Hence, this acts like a "transpose-free" method while still maintaining the smoothing properties of ADI. Algorithms are derived and convergence theory is provided.
Alternating Direction Implicit(ADI) methods have been in use since 1954 for the solution of both parabolic and elliptic partial differential equations. The convergence of these methods can be dramatically accelerated ...
详细信息
Alternating Direction Implicit(ADI) methods have been in use since 1954 for the solution of both parabolic and elliptic partial differential equations. The convergence of these methods can be dramatically accelerated when good estimates of the eigenvalues of the operator are available, However, in the case of computation on parallel computers, the solution of tridiagonal systems imposes an unreasonable overhead. We discuss methods to lower the overhead imposed by the solution of the corresponding tridiagonal systems. The proposed method has the same convergence properties as a standard ADI method, but all of the solves run in approximately the same time as the "fast" direction. Hence, this acts like a "transpose-free" method while still maintaining the smoothing properties of ADI. Algorithms are derived and convergence theory is provided.
Alternating Direction Implicit(ADI)methods have been in use since 1954 for the solution of both parabolic and elliptic partial differential *** convergence of these methods can be dramatically accelerated when good es...
详细信息
Alternating Direction Implicit(ADI)methods have been in use since 1954 for the solution of both parabolic and elliptic partial differential *** convergence of these methods can be dramatically accelerated when good estimates of the eigenvalues of the operator are available,However,in the case of computation on parallel computers,the solution of tridiagonal systems imposes an unreasonable *** discuss methods to lower the overhead imposed by the solution of the corresponding tridiagonal *** proposed method has the same convergence properties as a standard ADI method,but all of the solves run in approximately the same time as the “fast” ***,this acts like a ”transpose-free” method while still maintaining the smoothing properties of *** are derived and convergence theory is provided.
暂无评论