|
|
Performance Analysis of IBM POWER8 Processors
Jelen, Jakub ; Kešner, Filip (referee) ; Jaroš, Jiří (advisor)
This paper describes the IBM Power8 system in comparison to the Intel Xeon processors, widely used in today’s solutions. The performance is not evaluated only on the whole system level but also on the level of threads, cores and a memory. Different metrics are demonstrated on typical optimized algorithms. The benchmarked Power8 processor provides extremely fast memory providing sustainable bandwidth up to 145 GB/s between main memory and processor, which Intel is unable to compete. Computation power is comparable (Matrix multiplication) or worse (N-body simulation, division, more complex algorithms) in comparison with current Intel Haswell-EP. The IBM Power8 is able to compete Intel processors these days and it will be interesting to observe the future generation of Power9 and its performance in comparison to current and future Intel processors.
|
|
|
Acceleration of Lattice-Boltzmann Algorithms for Bloodflow Modeling
Kompová, Radmila ; Kešner, Filip (referee) ; Jaroš, Jiří (advisor)
This thesis aims to explore possible implementations and optimizations of the lattice-Boltzmann method. This method allows modeling of fluid flow using a simulation of fictive particles. The thesis focuses on possible improvements of the existing tool HemeLB which is designed and optimized for bloodflow modeling. Several vectorization and paralellization approaches that could be included in this tool are explored. An application focused on comparing chosen algorithms including optimizations for the lattice-Boltzmann method was implemented as a part of the thesis. A group of tests focused on comparing this algorithms according to performance, cache usage and overall memory usage was performed. The best performance achieved was 150 millions of lattice site updates per second.
|
|
|
Continuous Integration System for the k-Wave Project
Nečas, Radek ; Kešner, Filip (referee) ; Jaroš, Jiří (advisor)
The main goal of this thesis is to describe the implementation of continuous integration into the k-Wave project. The thesis focuses primarily on the version written in the C/C++ language with the usage of the OpenMP library which typically runs on supercomputers. Accordingly, many of popular workflows and approaches ought to be adapted, a few more created. The outcome of the thesis is a complete solution with real and practical usage. The author provides design, tools selection, runtime environment administration and configuration for each one of the used services. Software implementation of the basic framework is used in order to utilize running tests on the supercomputers. Furthermore, the implementation of chosen types of regression and unit tests are performed. Realisation is based on Gitlab and Jenkis services that are running on separated Docker containers.
|
|
|
MRI Data Processing Acceleration on GPU
Kešner, Filip ; Nečas, Ondřej (referee) ; Polok, Lukáš (advisor)
This BSc Thesis was performed during a study stay at the Universita della Svizzera italiana, Swiss. The identification of trajectories of neuron fibres within the human brain is of great importance in many medical applications as the neural diagnostics, neuronavigation, treatment of epilepsy, surgical removal of tumors and etc. By using diffusion MRI-data as input, and by employing Monte-Carlo like methods, possible trajectories are generated and the most likely ones can be visualized. These can serve as input for advanced medical diagnosis and treatments. Due to the huge amount of data to be analyzed and many iterations, this is a time consuming process. For the purposes such as statistical analysis and comparsion over several datasets or several patients, computational time requirements are enourmous. Faster diagnosis can improve routine throughput and provide earlier treatment of illness. At this time, there exists only a very few implementations of neural tractography sof tware. For probabilistic neural tractography is the list of software even thiner. Today's implementations using standard serial CPU execution suffer from high time consumption. The goal is to provide an efficient implementation which makes use of GPGPUs and exploits parallelism in the method. For the GPU implementation, a comparsion of CUDA and OpenCL technologies will be provided, using the more suitable one.
|
|
|
Platform for Safe File Transfer among Mobile Devices
Pustka, Michal ; Kešner, Filip (referee) ; Minařík, Miloš (advisor)
The thesis deals with safe files transfer among mobile devices . The first part of this thesis is the analysis and review of current solutions for transferring files among mobile devices and complex security solutions for Android operating system. The second part describes the design and evaluation of several concepts for encrypted files transfer among groups of regular users and outside the company or institution . One of the concepts for file transfer among users was chosen for implementation on the Android platform . The last part of the thesis describes the problems that can occur during the implementation , their solution , and finally the verification and validation of the concept itself .
|
|
|
Design of Digital Circuits at Transistor Level
Kešner, Filip ; Šimek, Václav (referee) ; Vašíček, Zdeněk (advisor)
This work aims to design process of integrated circuits on the transistor level, specially using evolutionary algorithm. For this purpose it is necessary to choose reasonable level of abstraction during simulation, which is used for evaluation candidate solutions by fitness function. This simulation has to be fast enough to evaluate thousands of candidate solutions within seconds. This work discusses already used techniques for transistor level circuit design and it chooses useful parts for new design of faster and more reliable automated design process, which would be able to design complex logic circuits. The thesis also discusses implementation of this system and used approach with regard to encountered problems in transistor-level circuit design and optimization by evolution.
|
|
|
Continuous Integration System for the k-Wave Project
Nečas, Radek ; Kešner, Filip (referee) ; Jaroš, Jiří (advisor)
The main goal of this thesis is to describe the implementation of continuous integration into the k-Wave project. The thesis focuses primarily on the version written in the C/C++ language with the usage of the OpenMP library which typically runs on supercomputers. Accordingly, many of popular workflows and approaches ought to be adapted, a few more created. The outcome of the thesis is a complete solution with real and practical usage. The author provides design, tools selection, runtime environment administration and configuration for each one of the used services. Software implementation of the basic framework is used in order to utilize running tests on the supercomputers. Furthermore, the implementation of chosen types of regression and unit tests are performed. Realisation is based on Gitlab and Jenkis services that are running on separated Docker containers.
|
|
|
Acceleration of Lattice-Boltzmann Algorithms for Bloodflow Modeling
Kompová, Radmila ; Kešner, Filip (referee) ; Jaroš, Jiří (advisor)
This thesis aims to explore possible implementations and optimizations of the lattice-Boltzmann method. This method allows modeling of fluid flow using a simulation of fictive particles. The thesis focuses on possible improvements of the existing tool HemeLB which is designed and optimized for bloodflow modeling. Several vectorization and paralellization approaches that could be included in this tool are explored. An application focused on comparing chosen algorithms including optimizations for the lattice-Boltzmann method was implemented as a part of the thesis. A group of tests focused on comparing this algorithms according to performance, cache usage and overall memory usage was performed. The best performance achieved was 150 millions of lattice site updates per second.
|
|
|
Performance Analysis of IBM POWER8 Processors
Jelen, Jakub ; Kešner, Filip (referee) ; Jaroš, Jiří (advisor)
This paper describes the IBM Power8 system in comparison to the Intel Xeon processors, widely used in today’s solutions. The performance is not evaluated only on the whole system level but also on the level of threads, cores and a memory. Different metrics are demonstrated on typical optimized algorithms. The benchmarked Power8 processor provides extremely fast memory providing sustainable bandwidth up to 145 GB/s between main memory and processor, which Intel is unable to compete. Computation power is comparable (Matrix multiplication) or worse (N-body simulation, division, more complex algorithms) in comparison with current Intel Haswell-EP. The IBM Power8 is able to compete Intel processors these days and it will be interesting to observe the future generation of Power9 and its performance in comparison to current and future Intel processors.
|
|
|
MRI Data Processing Acceleration on GPU
Kešner, Filip ; Nečas, Ondřej (referee) ; Polok, Lukáš (advisor)
This BSc Thesis was performed during a study stay at the Universita della Svizzera italiana, Swiss. The identification of trajectories of neuron fibres within the human brain is of great importance in many medical applications as the neural diagnostics, neuronavigation, treatment of epilepsy, surgical removal of tumors and etc. By using diffusion MRI-data as input, and by employing Monte-Carlo like methods, possible trajectories are generated and the most likely ones can be visualized. These can serve as input for advanced medical diagnosis and treatments. Due to the huge amount of data to be analyzed and many iterations, this is a time consuming process. For the purposes such as statistical analysis and comparsion over several datasets or several patients, computational time requirements are enourmous. Faster diagnosis can improve routine throughput and provide earlier treatment of illness. At this time, there exists only a very few implementations of neural tractography sof tware. For probabilistic neural tractography is the list of software even thiner. Today's implementations using standard serial CPU execution suffer from high time consumption. The goal is to provide an efficient implementation which makes use of GPGPUs and exploits parallelism in the method. For the GPU implementation, a comparsion of CUDA and OpenCL technologies will be provided, using the more suitable one.
|
guest ::
