OpenMP-based parallelization on an MPCore multiprocessor platform – A performance and power analysis

Chair for Electrical Engineering and Computer Systems, RWTH Aachen University, Schinkelstraße 2, 52062 Aachen, Germany
Journal of Systems Architecture (Impact Factor: 0.69). 11/2008; DOI: 10.1016/j.sysarc.2008.04.001
Source: DBLP

ABSTRACT In this contribution, the potential of parallelized software that implements algorithms of digital signal processing on a multicore processor platform is analyzed. For this purpose various digital signal processing tasks have been implemented on a prototyping platform i.e. an ARM MPCore featuring four ARM11 processor cores. In order to analyze the effect of parallelization on the resulting performance-power ratio, influencing parameters like e.g. the number of issued program threads have been studied. For parallelization issues the OpenMP programming model has been used which can be efficiently applied on C-level. In order to elaborate power efficient code also a functional and instruction level power model of the MPCore has been derived which features a high estimation accuracy. Using this power model and exploiting the capabilities of OpenMP a variety of exemplary tasks could be efficiently parallelized. The general efficiency potential of parallelization for multiprocessor architectures can be assembled.


Available from: Jörg Brakensiek, Mar 14, 2014
  • [Show abstract] [Hide abstract]
    ABSTRACT: High energy consumption is one of the key issues of cloud computing systems. Incoming jobs in cloud computing environments have the nature of randomness, and compute nodes have to be powered on all the time to await incoming tasks. This results in a great waste of energy. An energy-saving task scheduling algorithm based on the vacation queuing model for cloud computing systems is proposed in this paper. First, we use the vacation queuing model with exhaustive service to model the task schedule of a heterogeneous cloud computing system. Next, based on the busy period and busy cycle under steady state, we analyze the expectations of task sojourn time and energy consumption of compute nodes in the heterogeneous cloud computing system. Subsequently, we propose a task scheduling algorithm based on similar tasks to reduce the energy consumption. Simulation results show that the proposed algorithm can reduce the energy consumption of the cloud computing system effectively while meeting the task performance.
    Tsinghua Science & Technology 02/2015; 20(1):28-39. DOI:10.1109/TST.2015.7040511
  • [Show abstract] [Hide abstract]
    ABSTRACT: As chip multiprocessors have been widely adopted in embedded systems, achieving both high performance and low power consumptions of parallel applications becomes challenging. In order to meet these requirements, it is crucial for developers to analyze the performance and energy consumption of parallel applications. In this paper, we propose a tool for profiling and optimizing the performance and energy consumption of OpenMP applications (energy PROfiler and analyzer for OpenMP: ePRO-OMP). The main advantage of ePRO-OMP is that it can analyze both the performance and energy consumption of each parallel region of an OpenMP application, which can help developers find the bottleneck of parallel applications in detail.
    01/2011; 6(5).
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Future multimodal mobile platforms are expected to require high interactivity in their applications and user interfaces. Until now, mobile devices have been designed to remain in a stand-by state until the user actively turns it on in the interact ion sense. The motivation for this approach has been battery conservation. Imaging is a versatile sensing modality that can enable context recognition, unobtrusively predicting the user's interaction needs and directing the computational resources accordingly. However, vision-based always-on functionalities have been impractical in battery-powered devices, since their requirements of computational power a d energy make their use unattainable for extended periods of time. Vision-based app lications can benefit from the addition of interact ive stages that, properly designed, can reduce the complexity of the me thods utilizing user fee dback and collaboration, resulting in a system that balances computat ional throughput and energy efficiency. The usability of user interfaces critically rests on their latency. However, an always-on sensing platform needs a careful balance with the po wer consumption demands . Improving re activeness when designing for highly intera ctive vision-based interfaces ca n be achieved by reducing the number of operations that the application processor needs to exec ute, deriving the most expensive tasks to accelerators or specific processors. In this context, this thesis focuses on inves tigating and surveying enab lers and solutions for vision-based interactivity on mobi le devices. The thesis explor es the development of new user interaction methods by analyzing and comparing means to reach interactivity, high performance, low latency and energy efficiency. The researched techniques, ranging from mobile GPGPU and dedicated sensor processing to reconfigurable image processors, provide understanding on designing for future mobile platforms.
    12/2014, Degree: Doctor of Science, Supervisor: Olli Silvén