About NS3 Simulation Projects

Computing middlewares play a vital role for abstracting complexities of backend resources by providing a seamless access to heterogeneous execution management services. Scientific communities are taking advantage of such technologies to focus on science rather than dealing with technical intricacies of accessing resources. Multi-disciplinary communities often bring dynamic requirements which are not trivial to realize.

Specifically, to attain massivley parallel data processing on supercomputing resources which require an access to large data sets from widely distributed and dynamic sources located across organizational boundaries. In order to support this abstract scenario, we bring a combination that integrates UNICORE middleware and the Global Federated File System. Furthermore, the paper gives architectural and implementation perspective of UNICORE extension and its interaction with Global Federated File System space through computing, data and security standards.

The need for improved utilization of existing system assets and energy sources, as well as the smooth incorporation of new technologies (such as electric vehicles) into the grid, has prompted the participation of small power consumers and generators in the energy markets. A problem of such scale however cannot be managed in a centralized manner in its full detail. This paper examines the idea of a decentralized approach in clearing the energy market. A general framework for the problem decomposition and its distributed solution is presented and analyzed.

A key point of interest in this work is the fundamental question of how far decomposition may be pursued for a given system, while still achieving reasonable convergence properties. The corresponding optimization problem is formulated and solved through a parallel implementation of the alternating direction method of multipliers (ADMM). A thorough investigation of its convergence properties is conducted, and through its coordination with an additional proximal based decomposition method, we improve its scalability characteristics.

This paper presents a new approach to the distribution system state estimation in wide-area networks. The main goal of this paper is to present a two-step procedure designed to accurately estimate the status of a large-scale distribution network, relying on a distributed measurement system in a multiarea framework. First of all, the network is divided into subareas, according to geographical and/or topological constraints and depending on the available measurement system. Then, in the first step of the estimation process, for each area, a dedicated estimator is used, exploiting all the measurement devices available on the field.

In the second step, data provided by local estimators are further processed to refine the knowledge on the operating conditions of the network. To improve the accuracy of the estimation results, correlation arising in the first step estimations has to be suitably evaluated and considered during the second step. Performed analysis shows that existing correlations can be included in the estimation process with very low data exchange among areas, thus involving minimum communication costs. Both first and second steps can be performed in a decentralized way and withparallel processing, thus leading to reduced overall execution times. Test results, obtained on the 123-bus IEEE test network and proving the goodness of the proposed method, are presented and discussed.

Many modern advanced biology experiments are carried on by Electron Microscope(EM) image analysis. Segmentation is one of the most important and complex steps in the process of image analysis. Previous ISBI contest results and related research show that Convolution Neural Network(CNN)has high classification accuracy in EM image segmentation. Besides it eliminates the pain of extracting complex features which’s indispensable for traditional classification algorithms. However CNN’s extremely time-consuming and fault vulnerability due to long time execution prevent it from being widely used in practice. In this paper, we try to address these problems by providing reliable high performance CNN framework for medial image segmentation.

Our CNN has light weighted user level checkpoint, which costs seconds when doing one checkpoint and restart. On the fact of lacking in platform diversity in current parallel CNN framework, our CNN system tries to make it general by providing distributed cross-platform parallelism implementation. Currently we have integrated Theano’s GPU implementation in our CNNsystem, and we explore parallelism potential on multi-core CPUs and many-core Intel Phi by testing performance of main kernel functions of CNN. In the future, we will integrate implementation son other two platforms into our CNN framework.

Multicore GPU, Intel MIC, and FPGA supplemental parallel processors have become widely implemented in High Performance Computing Clusters (HPCCs). In HPCCs, Computing nodes are assembled with these supplemental processors for specific research applications, images are applied to do the research. Since HPCC computing nodes require completely different design configuration from one day to the next, System Administrators are being challenged to verify that each of these computing images work correctly, in all needed applications. Due to the large cost in man-hours that are expended with manual testing of each computing node and the entire HPCC system for defects, there is a need for automated regression testing on parallel, distributed, and heterogeneous computing nodes.

Existing approaches at automated regression testing deals only with simple homogeneous HPCC topologies. What is needed is a regression testing technique to include heterogeneous HPCC topologies that deal with computing nodes containing supplemental GPUs, Intel MIC cards, FPGAs, etc. This paper presents a case-study to perform regression testing using Equivalence Class Partitioning (ECP) and Boundary Value testing techniques. The method has been employed to test HPCCs configured of heterogeneous computing nodes. More specifically, the computing nodes configured for this experiment include NVidia GPU and Intel MIC Xeon Phi cards deployed in HPCC clusters.