The authors have published earlier a parallel - distributed implementation method for the supervised training of feed-forward artificial neural networks using the Harmony Search algorithm. Such implementation was inte...
详细信息
Elastic data stream processing enables applications to query and analyze streams of real time data. This is commonly facilitated by processing the flow of the data streams using a collection of stream processing opera...
详细信息
ISBN:
(纸本)9781728123653
Elastic data stream processing enables applications to query and analyze streams of real time data. This is commonly facilitated by processing the flow of the data streams using a collection of stream processing operators which are placed in the cloud. However, the cloud follows a centralized approach which is prone to high latency delay. For avoiding this delay, we leverage on the fog computing paradigm which extends the cloud to the edge of the network. In order to design a stream processing solution for the fog, we first formulate an optimization problem for the placement of stream processing operators, which is tailored to fog computing environments. Then, we build a plugin (for stream processing frameworks) which solves the optimization problem periodically in order to support the dynamic resources of the fog. We evaluate this approach by performing experiments on an OpenStack testbed. The results show that our plugin reduces the response time and the cost by 31.5% and 8.8% respectively, compared to optimizing the placement of operators only upon initialization.
This paper presents the StreamGen load generator, which is targeted at distributed information flow applications. These include the event streaming services used in wide-area publish/subscribe systems or in operationa...
详细信息
ISBN:
(纸本)0769521975
This paper presents the StreamGen load generator, which is targeted at distributed information flow applications. These include the event streaming services used in wide-area publish/subscribe systems or in operational information systems, the data streaming services used in remote visualization or collaboration, and the continuous data streams occurring in download services. Running across heterogeneous distributed platforms, these services are implemented by computational component that capture, manipulate, and produce information streams and are linked via overlay topologies. StreamGen can be used to produce the distributed computational and communication loads imposed by these applications. Dynamic application behaviors can be created with mathematical specifications or with behavior traces collected from application-level traces. An interesting set of traces presented in this paper is derived from long-term observations of the FTP download patterns observed at the Linux mirror site being run by the CERCS research center at the Georgia Institute of Technology. Two different flow-based applications are created and evaluated with StreamGen. The first emulates the data streaming behavior in a distributed scientific collaboration, where a scientific simulation (i.e., a molecular dynamics code) produces simulation data sent to and displayed for multiple, interactive remote users. The second emulates portions of the event-streaming behavior of an operational information system used by a large U.S. corporation. Parametric studies with StreamGen's FTP traces applied to these applications are used to evaluate different load balancing strategies for the cluster machines manipulating these applications' data streams.
In an earlier work on the design of fine-grain, scalable classifiers for massively parallel computers, the technique of unifying cascaded networks has been demonstrated. This paper further examines the method adopted ...
详细信息
ISBN:
(纸本)0818682596
In an earlier work on the design of fine-grain, scalable classifiers for massively parallel computers, the technique of unifying cascaded networks has been demonstrated. This paper further examines the method adopted using a highly parallelprocessing architecture, entitled Unified Hierarchical Classifiers (UHC), based on the principles of Generalised Regression Neural Networks (GRNN). As with the GRNN, it has been shown that the resulting classification network can be implemented efficiently on general-purpose multiprocessor platforms without dedicated hardware for processor interconnections. Adding to this the structural simplicity, and the demonstrable potential for an effective distributed realisation on the Cray T3D, will make UHC an attractive classifier architecture in practical applications.
Aimed at the limitation of ILP exploitation and the fixed topology of control-flow NP, a new scheme of coarse-grain dataflow NP architecture-DynaNP is proposed BynaNP can not only improve the programmability of NP by ...
详细信息
ISBN:
(纸本)0769529097
Aimed at the limitation of ILP exploitation and the fixed topology of control-flow NP, a new scheme of coarse-grain dataflow NP architecture-DynaNP is proposed BynaNP can not only improve the programmability of NP by adopting control-flow structure in each processing Element (PE), but also effectively exploit the task-level parallelism by introducing data-flow model into the processing of multiple PEs. A mechanism of dynamic configurable processing path is also provided in BynaNP. An SDTPPS algorithm based on dynamic configurable processing path is proposed in DynaNP. The simulation results show that by using this algorithm the load of each PE can be balanced efficiently and the overall throughput of DynaNP will be improved.
With the growth of supercomputer's scale, the communication time during executing is increasing. This phenomenon arouses the architecture researchers' interests. In this paper, based on the fat-tree topology, ...
详细信息
One hundred and fifty-six papers were presented at the Thirdinternational Joint conference on Pattern Recognition. The individual sessions covered the following topics: Industrial applications;Feature Extraction and ...
详细信息
One hundred and fifty-six papers were presented at the Thirdinternational Joint conference on Pattern Recognition. The individual sessions covered the following topics: Industrial applications;Feature Extraction and Primitive Selection;Syntactic Methods in Pattern Analysis;Optical Character Recognition;Learning Algorithms and Sample Size;Line Drawing and Waveform processing;Interactive Pattern Analysis;Statistical Pattern Recognition Theory;Perceptual Modeling;Pattern Recognition Competition;General applications;Clustering;Linguistic applications and Natural Language processing;Theoretical Problems;Segmentation and Shape Encoding;Medical Image processing and Pattern Analysis;Picture Description and Scene Analysis;Speech Recognition and Data Compression;Remote Sensing;parallelprocessing and Two-Dimensional Digital Filtering;Edge, Line and Object Recognition;applications of Pattern Recognition Technique;Image Analysis and Texture;Data Base Computer Systems.
The major goal of dynamic load balancing is not primarily to equalize the load on the nodes of a parallel computing system, but to optimize the average response time of single requests or the throughput of all applica...
详细信息
The major goal of dynamic load balancing is not primarily to equalize the load on the nodes of a parallel computing system, but to optimize the average response time of single requests or the throughput of all applications in the system. Therefore it is often necessary not only to keep all processors busy and all processor ready queue lengths within the same range, but to avoid delays and inefficient computations caused by foreseeable but ignored data flow and precedence constraints between related tasks. We will present concepts for dynamic consideration of inter task dependencies within small groups of tasks and evaluate them observing real applications in a load balancing environment on a network of workstations. The concepts are developed from scheduling of single task graphs towards heterogeneous multi user operation scenarios.
Many new database applications require very large volumes of data. Mariposa is a data base system under construction at Berkeley responding to this need. Mariposa objects can be stored over thousands of autonomous sit...
详细信息
Many new database applications require very large volumes of data. Mariposa is a data base system under construction at Berkeley responding to this need. Mariposa objects can be stored over thousands of autonomous sites and on memory hierarchies with very large capacity. This scale of the system leads to complex query execution and storage management issues, unsolvable in practice with traditional techniques. We propose an economic paradigm as the solution. A query receives a budget which it spends to obtain the answers. Each site attempts to maximize income by buying and selling storage objects, and processing queries for locally stored objects. We present the protocols which underlie the Mariposa economy.
暂无评论