prof. Ing. Pavel Tvrdík, CSc.

2014

The proposed project is based on previous research results in the field of parallel and distributed computing systems. It is aimed at massively parallel computing, GPU computing, cluster computing, and global grid computing systems. More specifically, the project is going to focus 1) on the reseach of the architecture of the nondedicated cluster architecture with the focus on the algorithms for distributed task scheduling in such clusters, 2) on the design of efficient algorithms for crystal structure determination based on powder diffraction method on massively parallel GPU clusters, 3) on the parallelization of immune-system-inspired algorithms on GPU clusters, 4) on the design of efficient algorithms for data acquisition and visualization of very large sparse matrices mapped row-wise/column-wise on processors of massively parallel multiprocessor systems, and 5) on the reseach of heuristic algorithms for resource allocation in worldwide computing grids.

2011 - 2012

The project goal is to design and create a framework for storing large sparse matrices onto disk subsystems and their loading back. The framework will be included in SA-NCSM (symmetry adapted no-core shell model) solver.

2012 - 2013

Research and development within DMS project aims at creation and implementation of modular SW used for increase of effectiveness of data centres - decrease of energy consumption and considerable decrease of operational costs. As a consequence, effective utilization of available area of a data centre can be reached resulting in higher density of allocated IT technologies and overall increase of accessibility of services.

2010 - 2012

The proposed project extends the previous research and development in the area of non-dedicated clusters. The original system, called Clondike, was successfully designed and implemented within a project supported 3year GAČR grant. During his doctoral study, Martin Stava designed a unique architecture combining elements of SSI clusters with peer-to-peer (P2P) networks. The new architecture, named ONDC system, brings along a number of new requirements on key support mechanisms, such as security, task scheduling, communication methods, cluster forming and monitoring, distributed file system. The goal of this project is to integrate these mechanisms into the proposed architecture.

2012 - 2014

Algorithms for solving so called "Grand challenge problems" lead to huge data sets, typically organized as sparse matrices. This project addresses the research of effective and scalable algorithms and data structures for input/output operations on very large sparse matrices that due to their size must be stored and processed on massively parallel computers with tens or hundreds of thousands of processors. Such matrices consist of trillions of nonzero entries. The project focuses on research of new binary file formats for storing such matrices, on research of data structures and scalable algorithms for effective loading such matrices into massively parallel solvers, and on research of memory-effective formats for representation of such matrices in computer memory. Finally, the project also aims at research of effective and scalable algorithms for visualization of very large sparse matrices on massively parallel computers. Together with theoretical parts, the project involves verification of proposed algorithms and data structures on real massively parallel computers.

2011

Visualization of sparse matrices is mostly performed on desktop systems by programs such as Matlab or Mathematica. This approach cannot be used in case of large matrices that are spread across many nodes of a massively parallel supercomputer. The goal of this project is to develop efficient visualization techniques for large sparse matrices that could be easily integrated into an existing codes based on the MPI parallel programming model.

2007 - 2009

This project is focused on industrial research in the area of high-speed radio-relay microwave lines. The goal of this project is design and development of new technological solutions and approaches and design and development of the modern price-frinedly high-speed raido-relay microwave line with the implemented transmission capacity of 155 Mbits per second and featured by QAM modulation. The main project contribution constitutes in proposition of a new technological solution of a microwave line, which is different from those ones currently implemented by the resolver. It also brings better technological features, namely the higher transmission capacity. Czech Republic will be the preferred distribution area of this device.

2016

The project covers 3 topics in the software architecture of distributed ICT systems: 1) Integration of the current architecture of the non-dedicated cluster system Clondike with an implementation of a distributed log in Apach Cassandra with the aim to realize sophisticated management of process migration and security management of the whole cluster. 2) Research of optimal policies of trust establishment and selfishness limitations in ad-hoc and sensor networks with the aim to find the required mix of requirements for computational tasks and requirements for forwarding and transmission of data within the network. The result will serve as a base for developping a method that will take into account not only message routing and trust establishment among modules of the network, but efficiency with respect to power awareness as well. 3) Research of the extension of the concept of Software Defined Networks (SDN) to dynamic circuit switching in optical networks with dense wave multiplex (DWDM) and research of impact of dynamic insertion and deletion of DWDM channels on the monitored parameters of the optical route (especially the impact of various signal levels of added channels on the quality of the DWDM signal at various parts of the route) with the aim to provide more efficient services for network applications requiring direct access to the photonic network layer.