Big Data and HPC: Ecosystem and Convergence

Nowadays High Performance Computing (HPC) and Big Data are  considered  the most  important and challengeable areas  in computing systems , with academic and industrial approaches. HPC trend and ecosystem is boosted by international exascale projects, and the Big Data ecosystem with a supplementing cloud infrastructure.  Big Data analytics are becoming well founded and applications in science and engineering more compute-intensive. In turn, working with much larger volumes of data, requires a crucial involvement of High Performance Computing systems.  As a matter of fact, Big Data based applications needs High Performance Computing Systems, high-speed storage-to-compute fabrics and high performance processing like fast CPUs, computational accelerators, and high-speed I/O. Therefore, the convergence of Big Data analytics, simulations and HPC is crucially important.

This workshop is aimed to address, explore and exchange information on the state-of-the-art in high performance, Big Data and large scale computing systems, their use in modeling and simulation, their design, performance and use, their impact and the convergence of HPC and Big Data ecosystems.

Steering  Сommittee

  • Lucio Grandinetti, University of Calabria, Italy
  • Seyedeh Leili Mirtaheri, Kharazmi University, Iran
  • Reza Shahbazian, Shahid Beheshti University, Iran


Manish Parashar, Department of Computer Science, Rutgers University, USA
Big Data Challenges in Extreme Scale Science

Data-related challenges are quickly dominating computational and data-enabled sciences, and are limiting the potential impact of scientific application workflows enabled by current and emerging extreme scale, high-performance, distributed computing environments. These data-intensive application workflows involve dynamic coordination, interactions and data coupling between multiple application processes that run at scale on different resources, and with services for monitoring, analysis and visualization and archiving, and present challenges due to increasing data volumes and complex data-coupling patterns, system energy constraints, increasing failure rates, etc. In this talk I will explore some of these data challenges and investigate how solutions based on data sharing abstractions, managed data pipelines, data-staging service, and in-situ / in-transit data placement and processing can be used to address these data challenges at extreme scales. This research is part of the Data Spaces project at the Rutgers Discovery Informatics Institute.

Ali Ahmadi, Computer Engineering Department, K. N. Toosi University of Technology, Iran
CBIR on big data by use of deep learning

Learning effective feature representations and similarity measures are critical in the performance of a CBIR. Although various techniques have been proposed, it remains one of the most challenging problems in CBIR, which is mainly due to "semantic gap" issue that exists between low-level image pixels captured by machine and high-level semantic concepts perceived by human.
One of the most important advances in machine learning is known as "deep learning" that attempts to model high-level abstractions in data by employing deep architectures composed of multiple non-linear transformations. We have improved CBIR using the state-of-the-art deep learning techniques for learning feature representations and similarity measures.
One of the most important methods for object detection is RCNN (Regions with CNN features). The core idea of RCNN is to generate multiple object proposals, extract features from each proposal using a CNN, and then classify each candidate window with a category-specific linear SVM. We have presented the idea of RCNN and the method improved this idea including multi-stage and deformable deep CNN.
The other important methods for object detection are based on a DNN-based regression towards an object mask. In these methods we can define one or multi CNN to detect multi objects. We will present the idea of DNN-based regression and compare this idea with the idea of RCNN.
In summary, our goal in this research is to explain the following statements:
1.    Various architectures in deep neural networks
2.    RCNN method for object detection
3.    DNN-based regression for object detection.

Valerio Pascucci, Inaugural John R. Parks Endowed Chair of the University of Utah, Director, Center for Extreme Data Management Analysis and Visualization (CEDMAV), Faculty, Scientific Computing and Imaging Institute Professor, School of Computing, University of Utah
Extreme Data Management Analysis and Visualization for Exascale Supercomputers

Effective use of data management techniques for analysis and visualization of massive scientific data is a crucial ingredient for the success of any supercomputing center and cyberinfrastructure for data-intensive scientific investigation. In the progress towards exascale computing, the data movement challenges have fostered innovation leading to complex streaming workflows that take advantage of any data processing opportunity arising while the data is in motion.
In this talk I will present a number of techniques developed at the Center for Extreme Data Management Analysis and Visualization (CEDMAV) that allow to build a scalable data movement infrastructure for fast I/O while organizing the data in a way that makes it immediately accessible for analytics and visualization. In addition, I will present a topological analytics framework that allows processing data in-situ and achieve massive data reductions while maintaining the ability to explore the full parameter space for feature selection.
Overall, this leads to a flexible data streaming workflow that allows working with massive simulation models without compromising the interactive nature of the exploratory process that is characteristic of the most effective data analytics and visualization environment.

Pejman Lotfi-Kamran, School of Computer Science, Institute for Research in Fundamental Sciences, Iran
Iran’s National Grid Initiative: Objectives, Challenges, and Opportunities

Nowadays, almost every organization has an IT department to offer services essential for the core business of the organization. These services include Website, data storage, business analytics, etc. Date centers are the backbone of such services. As demand on the data centers varies over time, organizations need to significantly overprovision the capacity of their data centers to handle worse case scenarios. Grid computing is a several decades old idea to connect computing resources of various organizations and enable them to provision the resources based on the average usage instead of worse case.
In this paper, we introduce Iran’s national grid initiative to connect computing resources of universities and offer a unified access point of all computing resources across several universities to the students regardless of where they study. The power of computing resources varies vastly across different universities: while some universities have more computational power than they actually need, other universities do not even have the necessary equipment to serve their students with the necessary computational power to do their homework. Moreover, even if universities have the necessary hardware, they usually lack the software stack that enable them to efficiently manage the hardware and offer a seamless service to the students. In this paper (talk), we introduce Iran’s national grid initiative: a comprehensive effort to unify the computational power of several universities across the country along with an easy-to-use interface for the students to have the computation they want without the need to know where the computation is being carried out. We go over the challenges of having such a system and the solutions that we come up with to address the challenges.  

Seyedeh Leili Mirtaheri, Computer Engineering Dept, Kharazmi University, Iran
Dynamic load balancing in distributed exascale computing systems

According to exascale computing roadmap, the dynamic nature of new generation scientific problems needs an undergoing review in the static management of computing resources. Therefore, it is necessary to present a dynamic load balancing model to manage the load of the system, efficiently. Currently, the distributed exascale systems are the promising solution to support the scientific programs with dynamic requests to resources. In this work, we propose a dynamic load balancing mechanism for distributed controlling of the load in the computing nodes. The presented method overcomes the challenges of dynamic behavior in the next generation problems. The proposed model considers many practical parameters including the load transition and communication delay. We also propose a compensating factor to minimize the idle time of computing nodes. We propose an optimized method to calculate this compensating factor. We estimate the status of nodes and also calculate the exact portion of the load that should be transferred to perform the optimized load balancing. The evaluation results show significant improvements regarding the performance by proposed load balancing in compared with some earlier distributed load balancing mechanisms.

Mitsuo Yokokawa, Professor, Graduate School of System Informatics, Kobe University, Japan
Koji Morishita, Takashi Ishihara, Atsuya Uno, and Yukio Kaneda

Performance of DNS of canonical tubulence and some simulation results on the K computer

Large scale direct numerical simulations (DNSs) of incompressible homogeneous turbulence in a periodic box with the number of grid points up to 12288^3 were carried out on the K computer. The DNS code was parallelized by using Message Passing Interface (MPI) and OpneMP with two dimensional domain decomposition. Simulation results and its performace will be presented in the talk. And some other simulation results on the K computer will be also presented breifly.