Search

Тема выпускной квалификационной работы: «Разработка системы анализа качества Ð´Ð°Ð½Ð½Ñ‹Ñ Ð¼Ð¾Ñ€ÑÐºÐ¾Ð¹ сейсморазведки».Магистерская работа посвящена исследованию морской сейсморазведки как разделу разведочной геофизики. В работе изучаются алгоритмы и функции, которые применяются при анализе качества Ð´Ð°Ð½Ð½Ñ‹Ñ , Ð¿Ð¾Ð»ÑƒÑ‡ÐµÐ½Ð½Ñ‹Ñ Ð² Ñ Ð¾Ð´Ðµ испытаний морской сейсморазведки. Также в данной работе рассматриваются Ð¿Ð¾Ð´Ñ Ð¾Ð´Ñ‹ к анализу качества Ð´Ð°Ð½Ð½Ñ‹Ñ Ð¼Ð¾Ñ€ÑÐºÐ¾Ð¹ сейсморазведки, а также алгоритмы по улучшению этого качества.В Ñ Ð¾Ð´Ðµ работы изучаются библиотеки языка разработки Python, с помощью ÐºÐ¾Ñ‚Ð¾Ñ€Ñ‹Ñ Ð¼Ð¾Ð¶Ð½Ð¾ реализовать все изученные ранее алгоритмы, а также визуализировать Ð¸Ñ Ñ€ÐµÐ·ÑƒÐ»ÑŒÑ‚Ð°Ñ‚Ñ‹ в виде Ð¸Ð½Ñ‚ÐµÑ€Ð°ÐºÑ‚Ð¸Ð²Ð½Ñ‹Ñ Ð³Ñ€Ð°Ñ„Ð¸ÐºÐ¾Ð².В результате проделанной работы реализована система анализа качества Ð´Ð°Ð½Ð½Ñ‹Ñ Ð¼Ð¾Ñ€ÑÐºÐ¾Ð¹ сейсморазведки на языке Python. Затем проведено её модульное и системное тестирование, описаны её преимущества, недостатки и направления дальнейшего её развития., The topic of the final qualifying work: "Development of a system for analyzing the quality of marine seismic data".The master's thesis is devoted to the study of marine seismic exploration as a section of exploration geophysics. The paper studies algorithms and functions that are used in analyzing the quality of data obtained during tests of marine seismic exploration. This paper also discusses approaches to analyzing the quality of marine seismic data, as well as algorithms to improve this quality.In the course of the work, Python development language libraries are being studied, with the help of which it is possible to implement all previously studied algorithms, as well as visualize their results in the form of interactive graphs.As a result of the work done, a system for analyzing the quality of marine seismic data in Python has been implemented. Then its modular and system testing was carried out, its advantages, disadvantages and directions of its further development are described.

As modern in-memory database systems achieve higher and higher processing speeds, the performance of memory becomes an increasingly limiting factor. Although there has been significant progress, the bottleneck only has shifted. While earlier systems were optimized for memory latencies, current systems are rather affected by the limited memory bandwidth. Query compilation is a proven technique to address bandwidth limitations. It translates queries via Just-In-Time compilation to native programs for the target hardware. The compiled queries execute with very high efficiency and only with a bare minimum of communication via memory. Despite these important improvements, the benefit of query compilation in certain scenarios is limited. On the one hand query compilers typically use standard compiler technology with relatively long compilation times. Therefore the overall execution time can be prolonged by the additional compilation time. On the other hand, not all emerging database technology is compatible with the approach. Query compilation uses a tuple-at-a-time processing style that departs from the column-at-a-time or vector-at- a-time approaches that in-memory systems typically use. Especially data-parallel processing techniques, e.g. SIMD or coprocessing-techniques, are challenging to use in combination with the approach. This work presents QCLab, a framework for query compilation on modern hardware platforms. The framework contains several new query compilation techniques that allow us to address the mentioned shortcomings and ultimately to extend the benefit of query compilation to new workloads and platforms. The techniques cover three aspects: compilation, communication, and processing. Together they serve as basis for building highly efficient query compilers. The techniques make efficient use of communication channels and of the large processing capacities of modern systems. They were designed for practical use and enable efficient processing, even when workload characteristics are challenging.

Recently, there is a growing need for distributed graph processing systems that are capable of gracefully scaling to very large datasets. In the mean time, in real-world applications, it is highly desirable to reduce the tedious, inefficient ETL (extract, transform, load) gap between tabular data processing systems and graph processing systems. Unfortunately, those challenges have not been easily met due to the intense memory pressure imposed by process-centric, message passing designs that many graph processing systems follow, as well as the separation of tabular data processing runtimes and graph processing runtimes.In this thesis, we explore the application of programming techniques and algorithms from the database systems world to the problem of scalable graph analysis. We first propose a bloat-aware design paradigm towards the development of efficient and scalable Big Data applications in object-oriented, GC enabled languages and demonstrate that programming under this paradigm does not incur significant programming burden but obtains remarkable performance gains (e.g., 2.5X). Based on the design paradigm, we then build Pregelix, an open source distributed graph processing system which is based on an iterative dataflow design that is better tuned to handle both in-memory and out-of-core workloads. As such, Pregelix offers improved performance characteristics and scaling properties over current open source systems (e.g., we have seen up to 15X speedup compared to Apache Giraph and up to 35X speedup compared to distributed GraphLab). Finally, we integrate Pregelix with the open source Big Data management system AsterixDB to offer users a mix of a vertex-oriented programming model and a declarative query language for richer forms of Big Graph analytics with reduced ETL pains.

Objective The purpose of this work is to optimize the processing of molecular dynamics (MD) trajectory data obtained for large biomolecular systems. Two popular software tools were chosen as the reference: the tng and the xdrfile libraries. Current implementation of tng algorithms and library is either fast or storage efficient and xdrfile is storage efficient but slow. Our aim was to combine speed and storage efficiency through the xdrfile’s code modification. Results Here we present libxtc, a ready-to-use library for reading MD trajectory files in xtc format. The effectiveness of libxtc is demonstrated for several biomolecular systems of various sizes (~ 2 × 104 to ~ 2 × 105 atoms). In sequential mode, the performance of libxtc is up to 1.8 times higher and 1.4 times lower than xdrfile and tng, respectively. In parallel mode, libxtc is about 3 and 1.3 times faster than xdrfile and tng. At the same time, MD data stored in the xtc format require about 1.3 times less disk space than those treated with the tng algorithm in the fastest reading mode, which is a noticeable saving especially when the MD trajectory is long and the number of atoms is large—this applies to most biologically relevant systems.

The emerging field of magnonics employs spin waves and their quanta, magnons, to implement wave-based computing on the micro- and nanoscale. Multi-frequency magnon networks would allow for parallel data processing within single logic elements whereas this is not the case with conventional transistor-based electronic logic. However, a lack of experimentally proven solutions to efficiently combine and separate magnons of different frequencies has impeded the intensive use of this concept. In this Letter, the experimental realization of a spin-wave demultiplexer enabling frequency-dependent separation of magnonic signals in the GHz range is demonstrated. The device is based on two-dimensional magnon transport in the form of spin-wave beams in unpatterned magnetic films. The intrinsic frequency-dependence of the beam direction is exploited to realize a passive functioning obviating an external control and additional power consumption. This approach paves the way to magnonic multiplexing circuits enabling simultaneous information transport and processing., 16 pages, 3 figures

В статье предложен подход к разработке математического обеспечения для поддержки процесса управления схемой реляционной СУБД, позволяющего учитывать статистику конкурентного доступа потока запросов к данным в иерархии памяти, используемой ядром СУБД. Отмечены недостатки существующих подходов, основанных на условной стоимости выполнения плана запроса, рассмотрен вопрос игнорирования издержек кооперативного доступа к данным в разделяемой памяти ЭВМ. Дается теоретикомножественное представление процесса обработки потока запросов, учитывающее недостатки существующих теоретикомножественных моделей. Предлагаемое представление обеспечивает учет фактических временных затрат для потока параллельно выполняемых запросов. Операции на высокоуровневом языке запросов предоставляются посредством множеств типовых операций доступа к данным в памяти. Разработана классификация операций доступа к памяти, позволяющая рассчитывать степень конкуренции при кооперативном выполнение запросов. Вводится формальное представление конкурентных запросов и условие выбора оптимального распределения данных в ходе функционирования базы данных на некотором промежутке времени. Полученные в работе результаты могут найти применение при разработке математического и программного обеспечения автономных систем управления базами данных, автоматизирующих управление физической схемой базы данных., The article proposes an approach to the development of mathematical support system for the relational database (database control system) schematic control process, allowing to take into account the statistics of competitive access to the data query flow in the memory used by the database control system engine by hierarchy. The articles notes shortcomings of existing approaches based on the conditional costs of the query plan and the issue of ignoring the costs related to cooperative access to the data contained in a shared computer memory. There is a settheoretic presentation of the query flow processing, which also takes into consideration the shortcomings of the existing settheoretic models. The proposed approach ensures the actual time costs calculations for the flow of concurrently executed queries. Highlevel query language operations are provided through a variety of typical access operations to the InMemory Data. The article presents a classification of the memory access operations, which allows to calculate the degree of competition during the cooperative query execution. This study also introduces a formal presentation of competitive queries and the conditions for choosing the optimal data distribution method during the database operation in a given period of time. The proposed settheoretic model allows us to calculate the memory segments, access to which leads to a competitive query. The results obtained in this paper can be used in the development of mathematical support systems and other software for the Autonomous database control systems that automate the management of the physical database scheme., №2(25) (2019)

Modern experimental facilities generate large amounts of data that should be processed, saved to hard disk and presented to the user as fast as possible. However, in-situ data analysis requires technical resources which are often not available. The existence of accessible high-speed networks allows to forward data processing and storage to a remote supercomputer centers and datacenters. These capabilities can be realized through the development of architectural solutions for effective data transmission trough a long-distance high-speed networks, data input/output and data distribution over computers and data storage systems. In this paper, we describe the results of investigations into the development of a software platform for parallel processing of intense experimental data-streams on ICMM UB RAS (Perm) and IMM UB RAS (Yekaterinburg) supercomputers, interconnected by a high-speed network. The reported studies was partially supported by RFBR, research project No. 14-07-96001-r_ural_a and by Program of UD RAS, project No 15-7-1-25.

Рассмотрены методы использования технологии облачных вычислений для разработки программного обеспечения, реализующего параллельную обработку распределенных данных. Рассмотрен метод параллельной реализации операции JOIN на основе принципа симметричного горизонтального распределения. Даны описания виртуальных программно-аппаратных комплексов, которые реализуют распараллеливание на уровне обработки запросов и на уровне операции JOIN средствами Microsoft Azure. Приведены экспериментальные данные, подтверждающие эффективность предложенного подхода., The methods of using cloud computing technology for the development of software that implements parallel processing of distributed data are considered. The method of parallel implementation of the JOIN operation based on the symmetric horizontal distribution principle is considered. There are descriptions of virtual software and hardware systems that implement parallelization at the level of processing requests and at the level of the JOIN operation using Microsoft Azure. Experimental data confirming the effectiveness of the proposed approach are presented., Международный научный журнал "Современные информационные технологии и ИТ-образование", № (2017)

This thesis is based on five publications related to the area of security assurance for hybrid clouds which were published at conferences or in journals by IEEE and Elsevier. Cloud computing is an ongoing research field that received an increasing attention in last several years, as new challenges arise in the cloud ecosystem on daily basis especially with the new hybrid cloud models coming to the scene. Meanwhile, the old challenges with regards to security, privacy and especially transparency haven't been comprehended or addressed properly to keep up with the technological momentum caused by the cloud paradigm. Sharing security sensitive information in a cloud environments has become a main obstacle due to the immense lack of transparency. Hence, this thesis addresses this challenges, in particular, transparency of cloud providers that for the given security objectives there are supporting measures in place. The thesis is especially concerned with the security and transparency with regards to security critical services especially with regards to hosting them in hybrid multi-layered and multi-tenant environments. To comprehend the complexity of hybrid cloud environments that can be composed of multiple layers and owned by multiple stakeholder we illustrate a composite multi-layer reference architecture model. The main objective of this model is to observe multidimensional critical infrastructure systems at individual levels from different viewpoints, namely those of multi-provider and multi-tenant, and different stakeholders. Furthermore, we analyze the challenges, objectives and requirements for deploying critical infrastructure services to cloud environments with regards to transparency and security. The challenges that we identified highlight the shortcoming of cloud providers to support transparency especially with regards to the hybrid cloud solutions. To overcome this gap this thesis propose a novel model for holistic security assurance assessment that addresses the interdependencies between both individual components and abstraction levels in hybrid cloud environments. The approach offers the ability to address each individual component of a cloud based infrastructure, regardless if it is a physical server, virtual container, or a high level service, in a structural manner by including all its interdependencies. The flexibility of the approach lies in the composite structural design of the security assurance assessment framework that adheres the Common Criteria and enhances it to achieve higher level of granularity when assessing services. Most importantly, unlike standard approaches for security assessment like certification or auditing our model offers continuous security assessment ability of hybrid cloud environments where we can have competitive cloud provider that deliver one single service. Lastly, our security assurance assessment model prevents the exposure of internal security sensitive information of a cloud provider via its novel security assurance assessment model that operates on abstracted security information sets. Furthermore, we propose a comprehensive life-cycle for designing, developing and deploying secure cloud services in line with standards, regulative compliance, and best practices. In addition, the proposed life-cycle integrates iterative security requirements engineering from high level objectives to security properties used for security validation entities through both development and production phase of cloud services. The life-cycle aligns and integrates the security assurance assessment model, by at the same time supporting it with security requirements, in the final production phase to enhance transparency.

Nowadays, we are witnessing the fast production of very large amount of data, particularly by the users of online systems on the Web. However, processing this big data is very challenging since both space and computational requirements are hard to satisfy. One solution for dealing with such requirements is to take advantage of parallel frameworks, such as MapReduce or Spark, that allow to make powerful computing and storage units on top of ordinary machines. Although these key-based frameworks have been praised for their high scalability and fault tolerance, they show poor performance in the case of data skew. There are important cases where a high percentage of processing in the reduce side ends up being done by only one node.In this paper, we present FP-Hadoop, a Hadoop-based system that renders the reduce side of MapReduce more parallel by efficiently tackling the problem of reduce data skew. FP-Hadoop introduces a new phase, denoted intermediate reduce (IR), where blocks of intermediate values are processed by intermediate reduce workers in parallel. With this approach, even when all intermediate values are associated to the same key, the main part of the reducing work can be performed in parallel taking benefit of the computing power of all available workers.We implemented a prototype of FP-Hadoop, and conducted extensive experiments over synthetic and real datasets. We achieved excellent performance gains compared to native Hadoop, e.g. more than 10 times in reduce time and 5 times in total execution time. HighlightsA novel approach for dealing with data skew in the reduce side of MapReduce.Parallel reducing of each key, using multiple reduce workers.Hierarchical execution of MapReduce jobs.Non-overwhelming reducing of intermediate data.

In the past, the majority of data analysis use cases was addressed by aggregating relational data. Since a few years, a trend is evolving, which is called “Big Data” and which has several implications on the field of data analysis. Compared to previous applications, much larger data sets are analyzed using more elaborate and diverse analysis methods such as information extraction techniques, data mining algorithms, and machine learning methods. At the same time, analysis applications include data sets with less or even no structure at all. This evolution has implications on the requirements on data processing systems. Due to the growing size of data sets and the increasing computational complexity of advanced analysis methods, data must be processed in a massively parallel fashion. The large number and diversity of data analysis techniques as well as the lack of data structure determine the use of user-defined functions and data types. Many traditional database systems are not flexible enough to satisfy these requirements. Hence, there is a need for programming abstractions to define and efficiently execute complex parallel data analysis programs that support custom user-defined operations. The success of the SQL query language has shown the advantages of declarative query specification, such as potential for optimization and ease of use. Today, most relational database management systems feature a query optimizer that compiles declarative queries into physical execution plans. Cost-based optimizers choose from billions of plan candidates the plan with the least estimated cost. However, traditional optimization techniques cannot be readily integrated into systems that aim to support novel data analysis use cases. For example, the use of user-defined functions (UDFs) can significantly limit the optimization potential of data analysis programs. Furthermore, lack of detailed data statistics is common when large amounts of unstructured data is analyzed. This leads to imprecise optimizer cost estimates, which can cause sub-optimal plan choices. In this thesis we address three challenges that arise in the context of specifying and optimizing data analysis programs. First, we propose a parallel programming model with declarative properties to specify data analysis tasks as data flow programs. In this model, data processing operators are composed of a system-provided second-order function and a user-defined first-order function. A cost-based optimizer compiles data flow programs specified in this abstraction into parallel data flows. The optimizer borrows techniques from relational optimizers and ports them to the domain of general-purpose parallel programming models. Second, we propose an approach to enhance the optimization of data flow programs that include UDF operators with unknown semantics. We identify operator properties and conditions to reorder neighboring UDF operators without changing the semantics of the program. We show how to automatically extract these properties from UDF operators by leveraging static code analysis techniques. Our approach is able to emulate relational optimizations such as filter and join reordering and holistic aggregation push-down while not being limited to relational operators. Finally, we analyze the impact of changing execution conditions such as varying predicate selectivities and memory budgets on the performance of relational query plans. We identify plan patterns that cause significantly varying execution performance for changing execution conditions. Plans that include such risky patterns are prone to cause problems in presence of imprecise optimizer estimates. Based on our findings, we introduce an approach to avoid risky plan choices. Moreover, we present a method to assess the risk of a query execution plan using a machine-learned prediction model. Experiments show that the prediction model outperforms risk predictions which are computed from optimizer estimates., In der Vergangenheit wurde die überwiegende Mehrheit der Datenanalyseanwendungen durch die Aggregation von relationalen Daten abgedeckt. Seit einigen Jahren entwickelt sich ein Trend der “Big Data” genannt wird und der große Auswirkungen auf den Bereich der Datenanalyse hat. Im Vergleich zu bisherigen Analyseanwendungen, werden nun wesentlich größere Datenmengen mit deutlich aufwändigeren und vielfältigeren Analysemethoden wie zum Beispiel Techniken der Informationsextraktion, des Data Minings, und Verfahren des maschinellen Lernens ausgewertet. Dabei werden auch Daten in die Analyse einbezogen, die weniger stark oder überhaupt nicht strukturiert sind. Die Veränderungen der Eigenschaften von Datenanalyseanwendungen wirken sich auch auf die Anforderungen an Systeme zur Datenverarbeitung aus. Aufgrund des gestiegenen Datenvolumens und des durch komplexere Analyseverfahren deutlich höheren Berechnungsaufwands müssen Daten massiv parallel verarbeitet werden. Die gestiegene Vielfalt von Analyseverfahren und die geringere Struktur der Daten erfordern häufig den Einsatz von benutzerdefinierten Funktionen und Datenstrukturen. Viele traditionelle Datenbanksysteme sind nicht flexibel genug, um diesen Anforderungen gerecht zu werden. Deshalb gibt es ein großes Interesse an neuen Programmierabstraktionen mit denen komplexe und parallele Datenanalyseanwendungen spezifiziert und effizient ausgeführt werden können. Der Erfolg der Anfragesprache SQL hat die Vorzüge von deklarativer Anfragespezifikation, wie zum Beispiel Optimierungspotenzial und Benutzerfreundlichkeit, deutlich gezeigt. Heute nutzt nahezu jedes relationale Datenbanksystem einen Anfrageoptimierer der deklarative Anfragen in physische Ausführungspläne übersetzt. Kostenbasierte Optimierer sind in der Lage aus Milliarden von möglichen Plänen einen effizienten Plan auszuwählen. Allerdings lassen sich traditionelle Optimierungsmethoden nicht ohne weiteres in Systeme integrieren, die neuartige Anwendungsfälle von Datenanalyse unterstützen wollen. Zum Beispiel kann der Einsatz von benutzerdefinierten Operationen das Optimierungspotenzial sehr stark reduzieren. Darüberhinaus sind selten detaillierte Datenstatistiken verfügbar, wenn große unstrukturierte Datensätze analysiert werden. Fehlende Statistiken haben häufig ungenaue Kostenschätzungen des Optimierers und somit die Auswahl von suboptimalen Ausführungsplänen zur Folge. In dieser Arbeit adressieren wir drei Herausforderungen im Kontext der Spezifikation und Optimierung von parallelen Datenanalyseprogrammen mit benutzerdefinierten Funktionen. Zunächst stellen wir ein paralleles Programmiermodell mit deklarativen Eigenschaften vor um Datenanalyseprogramme als Datenflußprogramme zu spezifizieren. In diesem Modell bestehen Datenverarbeitungsoperatoren aus einer systemeigenen Funktion zweiter Ordnung und einer benutzerdefinierten Funktion erster Ordnung. Ein kostenbasierter Optimierer übersetzt Datenflußprogramme, die in unserem Programmiermodell definiert wurden, in parallele Datenflüße. Unser Optimierer baut auf viele Techniken der relationalen Optimierung auf und überträgt sie in die Domäne von universellen parallelen Programmiermodellen. Zweitens präsentieren wir einen Ansatz zur Verbesserung der Optimierung von Datenflußprogrammen, die benutzerdefinierte Operatoren mit unbekannter Semantik enthalten. Wir identifizieren Eigenschaften von Operatoren und Bedingungen, um die Reihenfolge von benachbarten benutzerdefinierten Operatoren zu verändern ohne die Semantik eines Programms zu ändern. Wir zeigen wie diese Eigenschaften für benutzerdefinierte Operatoren vollautomatisch mit Hilfe von statischer Codeanalyse aus deren Quellcode extrahiert werden können. Mit unserem Ansatz können viele relational Optimierungen wie zum Beispiel die Optimierung der Reihenfolge von Filtern, Joins und Aggregationen emuliert werden ohne jedoch auf relationale Operatoren beschränkt zu sein. Drittens analysieren wir den Einfluß von sich verändernden Ausführungsbedingungen wie zum Beispiel variierenden Prädikatselektivitäten und verfügbaren Hauptspeichermengen auf die Laufzeit von relationalen Ausführungsplänen. Wir identifizieren Planeigenschaften, die deutliche Laufzeitschwankungen auslösen können. Im Fall von ungenauen Optimiererschätzungen können Pläne, die diese Eigenschaften enthalten, ein sehr großes Risiko darstellen. Wir präsentieren einen Ansatz, um die Auswahl von riskanten Plänen zu vermeiden. Darüberhinaus stellen wir eine Methode vor, um das Risiko von Ausführungsplänen mit Hilfe eines maschinell-gelernten Modells vorher zusagen. Unsere Evaluation zeigt, dass mit unserem Vorhersagemodell das Risikopotenzial eines Plans besser abgeschätzt werden kann als mit Hilfe eines kostenbasierten Optimierers.

This thesis demonstrates the development and the implementation of a real-time capable data processing and reconstruction algorithm for the ulftrafast X-ray scanner ROssendorf Fast Electron beam X-ray tomograph (ROFEX). This measuring system is built for non-invasive imaging of multiphase fluids. Thus, it requires a high scan rate of more than 1 kHz. This is achieved by an arrangement without mechanically rotating parts providing scan rates of up to 8 kHz. Current data processing is not suited to reconstruct a data stream of around 1,3 GB/s. Hence, visual inspection or active process feedback control is not possible yet. This work presents the design and implementation of real-time capable data processing providing a better usability and new fields of application for ROFEX. Therefore, the adjustment of data transfer from the measuring system to the reconstruction workstation as well as the implementation of a new software is essential. The application is implemented using a generic software pipeline consisting of distinct processing units exploiting data parallelism of NVIDIA GPUs with CUDA. Measurements on a NVIDIA GeForce GTX 1080 and NVIDIA Tesla K20c yield a reconstruction rate of more than 1 kHz, which is well-suited for an online application of ROFEX. Furthermore, a test system is introduced simulating an online capable detector. It shows the online visualization of reconstructed images as a first new application.

В статье рассматривается возможность использования энергоэффективного микропроцессора Epiphany для решения актуальной прикладной задачи - детекции лиц на изображении. Этот микропроцессор представляет собой многоядерную вычислительную систему с распределенной памятью, выполненную на одном кристалле. Из-за малой площади кристалла микропроцессор обладает существенными аппаратными ограничениями (в частности, он имеет всего 32 килобайта памяти на ядро), которые ограничивают выбор алгоритма и затрудняют его программную реализацию. Для детекции лиц адаптирован известный алгоритм, основанный на каскадном классификаторе, использующем LBP-признаки (Local Binary Patterns). Показано, что микропроцессор Epiphany, имеющий 16 ядер, может на этой задаче в 2,5 раза обогнать одноядерный процессор персонального компьютера той же тактовой частоты, при этом потребляя лишь 0,5 ватта электрической мощности. I t is studied the possibility of usage of energy-efficient Epiphany microprocessor for solving actual applied problem of face detection at still image. The microprocessor is a multicore system with distributed memory, implemented in a single chip. Due to small die area the microprocessor has significant hardware limitations (in particular it has only 32 kilobytes of memory per core) which limit the range of usable algorithms and complicate their software implementation. Common face-detection algorithm based on local binary patterns (LBP) and cascading classifier was adapted for parallel implementation. It is shown that Epiphany microprocessor having 16 cores can outperform single-core CPU of personal computer having the same clock rate by a factor of 2.5, while consuming only 0.5 watts of electric power. Сухинов Антон Александрович, к.ф. м.н., научный сотрудник, Сколковский институт науки и технологий (Сколково, Российская Федерация), soukhinov@gmail.com. Остроброд Георгий Борисович, программист ООО «СиВижинЛаб» (Таганрог, Российская Федерация), wdf.gost@gmail.com. A.A. Sukhinov, Skolkovo Institute of Science and Technology (Skolkovo, Russia), G.B. Ostrobrod, CVisionLab (Taganrog, Russia)

This thesis deals with the design and implementation of an optimizer and a transformer of relational queries. Firstly, the thesis describes the theory of the relational query compilers. Secondly, we present the data structures and algorithms used in the implemented tool. Finally, the important implementation details of the developed tool are discussed. Part of the thesis is the selection of used relational algebra operators and design of an appropriate input. Input of the implemented software is a query written in a XML file in the form of relational algebra. Query is optimized and transformed into physical plan which will be executed in the parallelization framework Bobox. Developed compiler outputs physical plan written in the Bobolang language, which serves as an input for the Bobox. Powered by TCPDF (www.tcpdf.org)

В предыдущих работах автором была предложена универсальная программная платформа, позволяющая реализовать известные алгоритмы метода группового учета аргументов, базисы, методы обучения и критерии селекции моделей. В статье приводится решение актуальной задачи увеличения производительности этой платформы. На основе проведенного обзора существующих архитектур вычислительных систем для параллельной обработки данных и программных систем индуктивного моделирования с поддержкой параллельных вычислений выработаны требования к подсистемам параллельных вычислений и управления памятью программной платформы. С использованием методологии объектно-ориентированного анализа и проектирования разработана объектно-ориентированная структура этих подсистем, приведены особенности их работы для каждой из указанных архитектур вычислительных систем. Производительность параллельной реализации комбинаторного алгоритма метода группового учета аргументов на базе программной платформы оценена экспериментально для многоядерных процессоров. In previous works the author has proposed a universal software framework that allows implementing the known algorithms of group method of data handling, model bases, training methods and model selection criteria. This paper introduces the solution of a topical problem of increasing the performance of the framework. Based on the review of existing computing architectures for parallel data processing and software systems for inductive modeling supporting parallel computations the author has worked out the requirements for the subsystems of parallel computing and memory management of the software framework. Using the methodology of object-oriented analysis and design the author developed the object-oriented structure of these subsystems and introduced the specifics of their operation on each of the mentioned computing architectures. The performance of the parallel implementation of the combinatorial group method of data handling algorithm on basis of the software framework was evaluated experimentally for multi-core processors.