Scheduling streaming applications in Data Stream Management Systems (DSMS) has been investigated for years. However, there lacks an intelligent system that is capable of monitoring application execution, modelling its...
详细信息
ISBN:
(纸本)9781538621295
Scheduling streaming applications in Data Stream Management Systems (DSMS) has been investigated for years. However, there lacks an intelligent system that is capable of monitoring application execution, modelling its resource usages, and then adjusting the scheduling plan under different sizes of inputs without requiring users' intervention. In this paper, we model the scheduling problem as a bin-packing variant and propose a heuristic-based algorithm to solve it with minimised inter-node communication. We also implement the D-Storm prototype to validate the efficacy and efficiency of our scheduling algorithm, by extending the Apache Storm framework into a self-adaptive MAPE (Monitoring, Analysis, Planning, Execution) architecture. The evaluation carried out on both synthetic and realistic applications proves that D-Storm outperforms the existing resource-aware scheduler and the default Storm scheduler by at least 16.25% in terms of the inter-node traffic reduction and yields a significant amount of resource savings through consolidation.
The use of massively parallel processors for high performance data management is moving rapidly into the commercial mainstream. This paper describes key features of the RISAM (Replicated ISAM) parallel data manager, w...
详细信息
The use of massively parallel processors for high performance data management is moving rapidly into the commercial mainstream. This paper describes key features of the RISAM (Replicated ISAM) parallel data manager, which is currently available on the C-DAC PARAM 8000 series of parallel supercomputers and on the Unisys U6000 symmetric multiprocessors, and which can be readily ported to other platforms. The design of RISAM makes it easy to migrate data management applications from a uniprocessor machine to a parallel architecture.
Maya is a platform for investigating the impact of different memory coherence protocols on parallel architectures. We present the implementations of several weak memory protocols, together with some new primitives ded...
详细信息
Maya is a platform for investigating the impact of different memory coherence protocols on parallel architectures. We present the implementations of several weak memory protocols, together with some new primitives dedicated to weak memories using Maya. The results of some user applications are summarized and the impact of weak memories on the efficiency of these parallel programs is discussed.
With the capability of communication, independent devices can be connected to share information and to perform a more complex task. In this paper, a distributed computing model suitable for industrial applications suc...
详细信息
ISBN:
(纸本)9781424417186
With the capability of communication, independent devices can be connected to share information and to perform a more complex task. In this paper, a distributed computing model suitable for industrial applications such as surveillance systems or mobile robots is introduced. The computing model is called the Dynamic distributed Computing (DDC) model. The dynamic join of distributed devices for sharing data together and processing tasks in parallel is an important characteristic of the proposed DDC model. For battery-enabled devices, the required electrical energy can be shared between the devices. Consequently, it not only can avoid the early termination of the devices due to running out of battery powers, but also can maintain the full functions provided by the complete devices for a steady period. The DDC model also supports a logical shared data space, through which information can be easily exchanged between devices. In addition, parallel programming over shared memory architecture is usually thought to be simpler than other methods. An environment supporting the DDC model, called D2MCE, is introduced in this paper. Experiment results show that good performance can be obtained by the DDC model, and the required electrical energy can be effectively shared by the participating devices.
Event processing has become the paradigm of choice in many monitoring and reactive applications. However, the understanding of events, their composition and level of abstraction, the style of processing and the qualit...
详细信息
ISBN:
(纸本)9781605586656
Event processing has become the paradigm of choice in many monitoring and reactive applications. However, the understanding of events, their composition and level of abstraction, the style of processing and the quality of service requirements vary drastically across application domains. We introduce the basic notions of event processing to create a common understanding, present the enabling technologies that are used for the implementation of event-based systems, survey a wide range of applications identifying their main features, and discuss open research issues. Copyright 2009 ACM.
The increase in performance of electronic computing systems may not be sustained in the face of interconnection bandwidth limitations. The potential of optics to overcome these limitations is discussed, with particula...
详细信息
ISBN:
(纸本)0780370961
The increase in performance of electronic computing systems may not be sustained in the face of interconnection bandwidth limitations. The potential of optics to overcome these limitations is discussed, with particular emphasis on planar optical packaging of free-space optics. Planar optics offers free-space optical functionality implemented with industry-standard manufacturing techniques. Our current research into applications of planar optics in digital communication and processing networks is presented.
This paper introduces a new mechanism for the exposure of large grain parallelism. The scheme performs lazy task creation;inlining all tasks provisionally and extracting parallelism from the inlined information later ...
详细信息
ISBN:
(纸本)0780342291
This paper introduces a new mechanism for the exposure of large grain parallelism. The scheme performs lazy task creation;inlining all tasks provisionally and extracting parallelism from the inlined information later on demand. However, unlike other mechanisms, the further task demand is satisfied by the next evaluation stream rather than retrospectively reversing the inlining decision of the current stream. The scheme is called lazy decomposition because decomposition itself is throttled rather than just the extraction of a task. Lazy decomposition makes the serial section clearly separated from the parallel section in an evaluation tree for a particular function, and this allows the serial section to adopt a sequential algorithm. The performance improvement is significant in divide-and-conquer applications by adoption of sequential algorithms.
As requirements for technical computing applications become more complex, engineers and scientists must solve problems of increasing computational intensity that frequently outstrip the capability of their own compute...
详细信息
ISBN:
(纸本)1424407281
As requirements for technical computing applications become more complex, engineers and scientists must solve problems of increasing computational intensity that frequently outstrip the capability of their own computers. Some are distributedapplications (also called coarse-grained or embarrassingly parallelapplications), where the same algorithm is independently executed over and over on different input parameters. Others consist of parallel (or fine-grained) applications, which contain interdependent tasks that exchange data during the execution of the application. This article introduces the distributed and parallel computing capabilities in The MathWorks distributed computing tools and provides examples of how these capabilities are applied to signal processingapplications.
A computational system based on a synchronous feedback neural network for the on-line event processing of a photon counting intensified CCD detector is presented. The hardware prototype, implemented by means of FPGA t...
详细信息
ISBN:
(纸本)0780342291
A computational system based on a synchronous feedback neural network for the on-line event processing of a photon counting intensified CCD detector is presented. The hardware prototype, implemented by means of FPGA technology, consists of 5x5 and is able to identify photon events against spurious and/or noise events. It shows a high level of flexibility, which is essential in the characterization phase of the detector. It allows to implement different kinds of neurons, that is having different output functions and internal architectures, and to run actual, as well as virtual, networks of neurons.
A comparative evaluation of distributed graphics processing over centralized graphics processing is presented. The evaluation is based on performance measurement software tools used in monitoring activities of graphic...
详细信息
A comparative evaluation of distributed graphics processing over centralized graphics processing is presented. The evaluation is based on performance measurement software tools used in monitoring activities of graphics functions and basic application packages. The factors measured are: frequency, elapsed real time, cpu time, I/O time, memory used, geometric model size, geometric entity picks. The performance statistics gathering software use special logic for extracting data pertinent to graphics. Key Words: GSP, GDS, GSR, emulation, geometric entity, intelligent terminal, statistics probes, benchmarks.
暂无评论