Nemanja Trifunovic

Guide to DataFlow Supercomputing: Basic Concepts, Case Studies, and a Detailed Example

This unique text/reference describes an exciting and novel approach to supercomputing in the DataFlow paradigm. The major advantages and applications of this approach are clearly described, and a detailed explanation of the programming model is provided using simple yet effective examples. The work is developed from a series of lecture courses taught by the authors in more than 40 universities across more than 20 countries, and from research carried out by Maxeler Technologies, Inc. Topics and features: presents a thorough introduction to DataFlow supercomputing for big data problems; reviews the latest research on the DataFlow architecture and its applications; introduces a new method for the rapid handling of real-world challenges involving large datasets; provides a case study on the use of the new approach to accelerate the Cooley-Tukey algorithm on a DataFlow machine; includes a step-by-step guide to the web-based integrated development environment WebIDE.

New benchmarking methodology and programming model for big data processing

Big data processing is becoming a reality in numerous real-world applications. With the emergence of new data intensive technologies and increasing amounts of data, new computing concepts are needed. The integration of big data producing technologies, such as wireless sensor networks, Internet of Things, and cloud computing, into cyber-physical systems is reducing the available time to find the appropriate solutions. This paper presents one possible solution for the coming exascale big data processing: a data flow computing concept. The performance of data flow systems that are processing big data should not be measured with the measures defined for the prevailing control flow systems. A new benchmarking methodology is proposed, which integrates the performance issues of speed, area, and power needed to execute the task. The computer ranking would look different if the new benchmarking methodologies were used; data flow systems would outperform control flow systems. This statement is backed by the recent results gained from implementations of specialized algorithms and applications in data flow systems. They show considerable factors of speedup, space savings, and power reductions regarding the implementations of the same in control flow computers. In our view, the next step of data flow computing development should be a move from specialized to more general algorithms and applications.

Paradigm Shift in Big Data SuperComputing: DataFlow vs. ControlFlow

The paper discusses the shift in the computing paradigm and the programming model for Big Data problems and applications. We compare DataFlow and ControlFlow programming models through their quantity and quality aspects. Big Data problems and applications that are suitable for implementation on DataFlow computers should not be measured using the same measures as ControlFlow computers. We propose a new methodology for benchmarking, which takes into account not only the execution time, but also the power and space, needed to complete the task. Recent research shows that if the TOP500 ranking was based on the new performance measures, DataFlow machines would outperform ControlFlow machines. To support the above claims, we present eight recent implementations of various algorithms using the DataFlow paradigm, which show considerable speed-ups, power reductions and space savings over their implementation using the ControlFlow paradigm.

An AppGallery for dataflow computing

This paper describes the vision behind and the mission of the Maxeler Application Gallery (AppGallery.Maxeler.com) project. First, it concentrates on the essence and performance advantages of the Maxeler dataflow approach. Second, it reviews the support technologies that enable the dataflow approach to achieve its maximum. Third, selected examples of the Maxeler Application Gallery are presented; these examples are treated as the final achievement made possible when all the support technologies are put to work together (internal infrastructure of the AppGallery.Maxeler.com is given in a follow-up paper). As last, the possible impact of the Application Gallery is presented and the major conclusions are drawn.

An Example Application: Fourier Transform

This chapter represents an example of accelerating the Cooley-Tukey algorithm with the Maxeler MAX3 machine and gives the results of the achieved acceleration. First, it explains the importance and usages of the Cooley-Tukey algorithm. Second, it gives mathematical explanation of the algorithm and algorithm’s pseudo code and explains different ways to implement the algorithm. The implementation with best time and memory complexity is explained in detail. Third, it explains how the algorithm has been accelerated using DataFlow engines. Fourth, it explains the experiments done to measure acceleration and present the results. The final results are presented as various graphs with explanations.

Cloud Deployment and Management of Dataflow Engines

Maxeler Technologies successfully commercialises high-performance computing systems based on dataflow technology. Maxeler dataflow computers have been deployed in a wide range of application domains including financial data analytics, geoscience and low-latency transaction processing. In the context of cloud computing steadily growing acceptance in new domains, we illustrate how Maxeler dataflow systems can be integrated and employed in a self-organising self-managing heterogeneous cloud environment.

Chapter Five - A Novel Infrastructure for Synergistic Dataflow Research, Development, Education, and Deployment: The Maxeler AppGallery Project.

Abstract This chapter presents the essence and the details of a novel infrastructure that synergizes research, development, education, and deployment in the context of dataflow research. To make it clearer to fundamental scientists, the essence of the approach is explained by referencing the results of the work of four different Nobel laureates. To make it clearer to research community, crucial details are presented in the form of a manual. Till this point, the development of dataflow applications was based on tools inherited from the controlflow environment. We here describe a set of tools developed from scratch with dataflow specifics in mind. These tools are not only tuned to the dataflow environment, but they are also tuned to synergize with each other, for the best possible performance in minimal time, counting from the moment when new researchers enter the dataflow arena, until the moment when they are able to deliver a quality code for maximal speed performance and minimal energy consumption. The effectiveness of the presented synergetic approach was measured empirically, using a group of students in a dataflow course. The measured results clearly indicate the superiority of the proposed approach in the following five domains: time to design, time to program, time to build, time to test, and speedup ratio.

Big Data Processing: Data Flow vs Control Flow (New Benchmarking Methodology)

Big Data processing is becoming a reality in numerous real-world applications. One very important area of research with a rapid growth of data volume is sensor networks. This article discusses the shift in the computing paradigm for Big Data problems and applications. We briefly introduce the Data Flow programming model and then focus on the new benchmarking methodology for Big Data processing. Big Data problems and applications that are suitable for implementation on Data Flow computers should not be measured using the same measures as Control Flow computers. We propose a new benchmarking methodology, which takes into account not only the execution time, but also the power and space, needed to complete the task. Recent research shows that if the Top 500 ranking was based on the new performance measures, Data Flow machines would outperform Control Flow machines. To support the above claims, we present some recent implementations of various algorithms using the Data Flow paradigm, which show considerable speed-ups, power reductions, and space savings over their implementation using the Control Flow paradigm.

DataFlow Systems: From Their Origins to Future Applications in Data Analytics, Deep Learning, and the Internet of Things

With the slowdowns in Dennard scaling and limited performance gain in multi-core scaling, we are witnesses of the high-performance computing shift to domain-specific hardware systems which empower big data and high-performance applications. Likewise, dataflow systems are experiencing a revival with both hardware and software approaches widely exploited. In our work, we give an overview of dataflow system origins and similar technologies such as systolic architecture whose principles are applied by some of today’s leading high-performance systems such as Multiscale dataflow Computing (MDC). In the second part, we highlight certain applications that could benefit from delegating critical processing to a MDC system. We emphasize algorithms and applications from data analytics, deep learning, and the Internet of Things (IoT), with a special focus on their execution within the cloud environment. We discuss the integration of software distributed dataflow systems such as Apache Spark with MDC systems, analyze design issues and challenges for implementation of deep neural networks using MDC, and how semantic-enabled IoT platforms and services could be improved by using MDC systems in order to become more effective. We expect that these selected case studies would motivate researchers to investigate engagement of hardware dataflow systems to support applications from other areas with similarly rigid requirements.

Introductory Overview on Implementation Tools

When programming the dataflow engines, developers need to change the way of thinking from the way of thinking used when programming CPUs where a program that gets executed in time is written to thinking about how to write a spatial recipe that will best configure the dataflow engines so that the data gets processed in space flowing through the configured devices.