Advances in object-oriented (OO) technology and distributedcomputing generate distributed object oriented software systems. Choosing an efficient design of such software is a multi-criteria decision problem. Performa...
详细信息
ISBN:
(纸本)9780780387355
Advances in object-oriented (OO) technology and distributedcomputing generate distributed object oriented software systems. Choosing an efficient design of such software is a multi-criteria decision problem. Performance is a key criterion that makes software developers be able to select the system that best fit their requirements and achieve longer design stability. Classical techniques of performance analysis are either unsuitable or unnatural to capture performance behavior of OO systems. In this paper, we present Performance-Based Model for distributed OO Software (DOOS) and a methodology to analyze and evaluate its performance. The new model evaluates the time cost of DOOS system considering the communication overheads while preserving the OO features such as encapsulations and inheritance.
In this paper, we consider the construction of a large scale wireless sensor network (WSN) by means of distributed network association with scalability. We design a network association mechanism that allows each route...
详细信息
ISBN:
(纸本)9781479932474;9780769550237
In this paper, we consider the construction of a large scale wireless sensor network (WSN) by means of distributed network association with scalability. We design a network association mechanism that allows each router to have its own addressing space for unique addressing of its child nodes based on a pre-initialized addressing tree. We also design a device selection algorithm that can locate routers throughout the whole deployment area using a minimum number of routers (i.e., minimizing the energy consumption). Thus, nodes can securely join the network through a sufficient number of neighboring routers, with the use of a uniquely allocated address in a distributed manner. Finally, the performance of the proposed scheme is evaluated by computer simulation, showing remarkable performance improvement over conventional schemes when applied to the construction of a large scale WSN.
Wireless sensor Networks (WSNs) pose novel challenges compared with traditional networks. To answer such challenges a new communication paradigm, data-centric communication, is emerging. One form of data-centric commu...
详细信息
ISBN:
(纸本)9781424417964
Wireless sensor Networks (WSNs) pose novel challenges compared with traditional networks. To answer such challenges a new communication paradigm, data-centric communication, is emerging. One form of data-centric communication is the publish/subscribe messaging system. Compared with other data-centric variants, publish/subscribe systems are common and wide-spread in distributedcomputing. Thus, extending publish/subscribe systems into WSNs will simplify the integration of sensor applications with other distributed applications. This paper describes MQTT-S [1], an extension of the open publish/subscribe protocol Message Queuing Telemetry Transport (MQTT) [2] to WSNs. MQTT-S is designed in such a way that it can be run on low-end and battery-operated sensor/actuator devices and operate over band width-constraint WSNs such as ZigBee-based networks. Various protocol design points are discussed and compared. MQTT-S has been implemented and is currently being tested on the IBM wireless sensor networking testbed [3]. Implementation aspects, open challenges and future work are also presented.
Separately, context-aware sensing and networked sensing systems have been fast progressing research domains. The emergence of wireless sensor networks (WSN) has introduced a way of implementing a distributed self-orga...
详细信息
Many self-configuration problems that occur in sensor networks, such as clustering or operator placement for in-network data aggregation, can be modeled as facility location problems. Unfortunately, existing distribut...
详细信息
ISBN:
(纸本)9783540730897
Many self-configuration problems that occur in sensor networks, such as clustering or operator placement for in-network data aggregation, can be modeled as facility location problems. Unfortunately, existing distributed facility location algorithms are hardly applicable to multi-hop sensor networks. Based on an existing centralized algorithm, we therefore devise equivalent distributed versions which, to our knowledge, represent the first distributed approximations of the facility location problem that can be practicably implemented in multi-hop sensor networks with local communication. Through simulation studies, we demonstrate that, for typical instances derived from sensor-network configuration problems, the algorithms terminate in only few communication rounds, the run-time does not increase with the network size, and, finally, that our implementation requires only local communication confined to small network neighborhoods. In addition, we propose simple extensions to our algorithms to support dynamic networks with varying link qualities and node additions and deletions. Using link quality traces collected from a real sensor network deployment, we demonstrate the effectiveness of our algorithms in realistic multi-hop sensor networks.
Stream processing is emerging to react to the changing business situations of real-time processing. The main aim of this paradigm is to deal with the huge volume of data in the format of information flows originating ...
详细信息
ISBN:
(纸本)9781538625880
Stream processing is emerging to react to the changing business situations of real-time processing. The main aim of this paradigm is to deal with the huge volume of data in the format of information flows originating from distributed devices. This consequently poses challenges to the scheduling problem in cloud data centers regarding the time-varying velocity of data ingesting and processing. In response to the uncertainties and complexities of streaming data, we propose a model-based scheduling scheme for stream processing systems, capturing the system behavior and providing an optimal allocation strategy to adapt to the changing work conditions. The proposed scheduling policy is implemented in Apache Storm, and micro-benchmarks with various shapes (e.g line, star, and diamond) were used in the evaluation. A topology that tracks trending topics on Twitter is also used, where the input is feeding with tweets in realtime. Experimental results show that the proposed solution can perform estimations that are well aligned with the system performance. The proposed scheduling policy achieves an improved performance with regards throughput and latency under varying ingesting rates.
Wireless sensor networks are often densely deployed for environmental monitoring applications. Collecting raw data from these networks can lead to excessive energy consumption. Thus using the spatial and temporal corr...
详细信息
ISBN:
(纸本)9783540730897
Wireless sensor networks are often densely deployed for environmental monitoring applications. Collecting raw data from these networks can lead to excessive energy consumption. Thus using the spatial and temporal correlations that exist between adjacent nodes we appoint a few as representative nodes that perform in-network aggregation. This reduces the total number of transmissions. Our distributed scheduling algorithm autonomously assigns a particular node to perform aggregation and reassigns schedules when network topology changes. These topology changes are detected using cross-layer information from the underlying MAC layer. We also present theoretical performance estimates and upper bounds of our algorithm and evaluate it by implementing the algorithm on actual sensor nodes, demonstrating an energy-saving of up to 80% compared to raw data collection.
distributed deep learning systems place stringent requirement on communication bandwidth in its model training with large volumes of input data under user-time constraint. The communications take place mainly between ...
详细信息
ISBN:
(纸本)9781538625880
distributed deep learning systems place stringent requirement on communication bandwidth in its model training with large volumes of input data under user-time constraint. The communications take place mainly between cluster of worker nodes for training data and parameter servers for maintaining a global trained model. For fast convergence the worker nodes and parameter servers have to frequently exchange billions of parameters to quickly broadcast updates and minimize staleness. Demand on the bandwidth becomes even higher with the introduction of dedicated GPUs in the computation. While rdMA-capable network has a great potential to provide sufficiently high bandwidth, its current use over TCP/IP or tied to particular programming models, such as MPI, limits its capability to break the bandwidth bottleneck. In this work we propose irdMA, an rdMA-based parameter server architecture optimized for high-performance network environment supporting both GPU- and CPU-based training. It utilizes native asynchronous rdMA verbs to achieve network line speed while minimizing the communication processing cost on both worker and parameter-server sides. Furthermore, irdMA exposes the parameter server system as a POSIX-compatible file API for convenient support of load balance and fault tolerance as well as its easy use. We have implemented irdMA at IBM's deep learning platform. Experiment results show that our design can help deep learning applications, including image recognition and language classification, to achieve near-linear improvement on convergence speed and training accuracy acceleration by using distributedcomputing resources. From the system perspective, irdMA can efficiently utilize about 95% network bandwidth of fast networks to synchronize models among distributed training processes.
The paper describes a new method for the synthesis of the application specific distributed control systems, constructed using the FPGA devices. The initial steps of the proposed control algorithm rely on the notion of...
详细信息
ISBN:
(纸本)9781467347211
The paper describes a new method for the synthesis of the application specific distributed control systems, constructed using the FPGA devices. The initial steps of the proposed control algorithm rely on the notion of a Petri net, which is an easy way to describe parallel processes. The subsequent steps of the algorithm consist in the decomposition of a given Petri net - with the use of a coloring algorithm - into a set of state machine type subnets. We think of each subnet as representing one parallel process. These subnets are then implemented independently in different FPGA devices. To ensure proper communication between all subnets, the entire control system uses a globally asynchronous locally synchronous (GALS) architecture with each subnet synchronized by the local clock signal. Global communication between components is buffer-based and uses additional signals, generated in a given subnet and distributed to the remaining ones.
暂无评论