|
|
Simulation of the Heat Diffusion with a Time-Varying Source on GPUs
Hála, Pavel ; Záň, Drahoslav (referee) ; Jaroš, Jiří (advisor)
This bachelor's thesis deals with the simulation of the heat transfer inside human tissue injected by an external time varying heat source. The proposed implemented simulation is based on a 4th order in space and 1st order in time finite-difference time domain method. First, a multithreaded CPU version was implemented. Subsequently, several GPU accelerated versions were implemented taking into account architecture aspect of the GPU. The experimental results showed that the fastest GPU kernel was the naive one using only the GPU global memory. Next, the usefulness of the Gauss-Seidel's method was investigated. The CPU implementation of the method was evaluated as usable because of being only 13% slower while saving up to 50% of memory resources. However, the GPU implementation was twice as slow as the naive version mainly due to shared memory size limits. The peak performance in terms of GFLOPS reached 32 and 135 on CPU and GPU, respectively. This corresponds to 10% and 9% of the theoretical potential of given architectures.
|
|
|
Development of Inverse Tasks Solved by Using the Optimizing Procedures and Large Number of Parallel Threads
Ondroušková, Jana ; Skarolek, Antonín (referee) ; Brestovič, Tomáš (referee) ; Horský, Jaroslav (advisor)
In metallurgy it is important to know a cooling efficiency of a product as well as cooling efficiency of working rolls to maximize the quality of the product and to achieve the long life of working rolls. It is possible to examine this cooling efficiency by heat transfer coefficients and surface temperatures. The surface temperature is hardly measured during the cooling. It is better to compute it together with heat transfer coefficient by inverse heat conduction problem. The computation is not easy and it uses estimated values which are verified by direct heat conduction problem. The time-consuming of this task can be several days or weeks, depends on the complexity of the model. Thus there are tendencies to shorten the computational time. This doctoral thesis considers the possible way of the computing time shortening of inverse heat conduction problem, which is the parallelization of this task and its transfer to a graphic card. It has greater computing power than the central processing unit (CPU). One computer can have more compute devices. That is why the computing time on different types of devices is compared in this thesis. Next this thesis deals with obtaining of surface temperatures for the computation by infrared line scanner and using of inverse heat conduction problem for the computing of the surface temperature and heat transfer coefficient during passing of a test sample under cooling section and cooling by high pressure nozzles.
|
|
|
A New Generation of an IPFIX Collector
Huták, Lukáš ; Žádník, Martin (referee) ; Wrona, Jan (advisor)
This master's thesis addresses processing of flow monitoring records from a point of view of an IPFIX collector. It analysis the current solution of the modular collector, which went through considerable historical development, and focuses on revealing its strengths and weaknesses. Based on acquired knowledge, a new collector is designed. The new solution, which significantly modifies individual components for processing of flow records, focuses on high throughput and adds missing functionalities. The document also compares performance of both generations and the new collector clearly dominates.
|
|
|
Diffusion Evolutionary Algorithm
Žundálek, Zbyněk ; Puš, Viktor (referee) ; Jaroš, Jiří (advisor)
This bachelor thesis deals with a parallelization of cellular evolutionary algorithms using OpenMP. The theoretical part of the thesis contains an introduction to evolutionary and genetic algorithms followed by the description of their parallel implementation on shared memory systems. This part is completed with the OpenMP key features analysis. The practical part of this thesis describes two possible implementations of a diffusion evolutionary algorithm; synchronous and asynchronous. The comparison of achievable performance of these two methods carried out on the N-Queen problem is provided in the experimental part of the thesis. The quality of found solutions is further examined with respect to the neighborhood size, topology and the replacement operator of the diffusion evolutionary algorithm.
|
|
|
The Parallelization of a Compilation and Linking Process
Koláček, Vojtěch ; Vašíček, Zdeněk (referee) ; Slaný, Karel (advisor)
This bachelor's thesis deals with the parallelization of a compilation and linking process. Various tools for the automatization of the compilation process are discussed. The main criteria is focused on the ability of parallelization of the compilation process. Three systems which are able to parallelize the compilation process have been implemented. The implemented systems cooperate with Microsoft C/C++ Compiler and Intel C++ Compiler.
|
|
|
Lifting Scheme Cores for Wavelet Transform
Bařina, David ; Kälviäinen, Heikki (referee) ; Sojka, Eduard (referee) ; Zemčík, Pavel (advisor)
Práce se zaměřuje na efektivní výpočet dvourozměrné diskrétní vlnkové transformace. Současné metody jsou v práci rozšířeny v několika směrech a to tak, aby spočetly tuto transformaci v jediném průchodu, a to případně víceúrovňově, použitím kompaktního jádra. Tohle jádro dále může být vhodně přeorganizováno za účelem minimalizace užití některých prostředků. Představený přístup krásně zapadá do běžně používaných rozšíření SIMD, využívá hierarchii cache pamětí moderních procesorů a je vhodný k paralelnímu výpočtu. Prezentovaný přístup je nakonec začleněn do kompresního řetězce formátu JPEG 2000, ve kterém se ukázal být zásadně rychlejší než široce používané implementace.
|
|
|
Optimization of KPI Processing
Šulc, Ondřej ; Bartík, Vladimír (referee) ; Hynek, Jiří (advisor)
This thesis deals with the optimization of data processing from IoT sensors of smart cities into the form of key performance indicators (abbr. KPI). KPIs are a mean of monitoring a large amount of data and expressing the status of performance factors affecting the prosperity of the entire city. Data processing in this form is a computationally demanding process, but it consists of a large number of mutually independent calculations. Therefore the goal of this thesis was to perform optimization using parallelization. In parallel processing, calculations can be divided between multiple threads, enabling all available computing resources (CPU cores) to be fully used. This concept was practically implemented in the Smart City project of Logimic company. However, the project is built on the Node.js platform, and when using parallelization there are complications with the use of libraries for object-relational mapping (abbr. ORM). ORM libraries on the Node.js platform are not always ready to work in a parallel environment. This problem is solved by creating a separate instance of the used library for each parallel thread. The thesis focuses on reducing the overhead associated with this and also on the correct distribution of work between parallel threads so that all cores are used equally. The results of this work prove that optimizing IoT data processing using parallelization leads to a significant speedup that conforms to Amdahl's law, as overhead problems can be reduced to a negligible minimum.
|
|
|
Research of Methods for Significant Accereation of Parameter Estimation of Simulation Models
Appel, Martin ; Opluštil, Vladimír (referee) ; Křivánek, Václav (referee) ; Grepl, Robert (advisor)
The thesis focuses on achieving a significant speed-up in the estimation of simulation model parameter values. This is achieved through the appropriate choice of a solver that is computationally less demanding, but at the same time has an acceptable error in the range of parameter values, through the use of distribution of computations on parallel threads, heuristic methods for reducing the space of parameter values, and modification of optimization methods. In this thesis, the research related to the objectives of the thesis is first discussed, then the exact objectives of this thesis are stated and the implementation of each objective is described in separate chapters. The results of this thesis include a tool used to analyze solver choice, an analysis of parallel thread efficiency, a parallel simulation distribution tool, nine modified optimization methods, and a new tool for estimating parameter values. Finally, the results obtained are evaluated.
|
|
|
CPU Rendering of Large Volumetric Data
Svoboda, Jan ; Vlnas, Michal (referee) ; Španěl, Michal (advisor)
This thesis deals with design and implementation of a system that allows displaying large volumetric data in real time on the CPU of a conventional computer. The thesis aims to solve two biggest problems. Firstly, it aims to solve the problem with rendering itself, where this amount of data often cannot be placed into the main memory of a target computer. Secondly, it aims to solve the problem of storing of this data, where, in the case of large datasets, storing them in the storage of a target computer may not be desirable. The proposed solution contains two applications -- the server one and the client one. The server part is used as a remote storage of volumetric data that is provided to the client application in small blocks and in different qualities. The client application renders this data by the ray casting method and, according to the created strategies, performs loading and storing of required blocks in the local memory. In order to achieve high performance, the client application was implemented with an emphasis on parallelization of the main processes. The resulting system allows a user to display large datasets stored on a server's storage and to manage the datasets using a simple graphical user interface.
|
|
|
Acceleration of 2D Wavelet transform on parallel architectures
Kula, Michal ; Schier, Jan (referee) ; Sojka, Eduard (referee) ; Zemčík, Pavel (advisor)
I přesto, že byla 2D diskrétní vlnková transformace předmětem řady rozsáhlých studií, některé aspekty této problematiky byly doposud opomíjeny. Mezi takové aspekty lze zařadit techniky pro výpočet této transformace se zaměřením na vyvažování synchronizací, aritmetických instrukcí a využití paměti pro různé architektury. Tato práce ukazuje několik nových metod výpočtu této transformace s různě nastaveným vybalancováním těchto operací. Tyto metody jsou detailně popsány a jejich chování je vyhodnoceno na několika grafických adaptérech za použití GPGPU, zpracování pomocí grafické pipeliny a vícejádrových procesorů pomocí OpenMP.
|
guest ::
