FPM Based Partitioning and Assignment Algorithm for Data Parallel Applications on Heterogeneous Platforms

2022-7-01
Alasmar, Mahmoud
Advances in modern computing devices and applications created the challenge of efficient utilization of resources in satisfying the requirements of running applica- tions. The present work aims to find an efficient workload distribution algorithm for data parallel applications of type single program multiple data (SPMD) running on a heterogeneous computing platform. We first consider a discrete functional perfor- mance model (FPM) that integrates processing speed and capacity of processing ele- ments with the size of the computational task. We then develop a mathematical model and propose an appropriate heuristic mapping algorithm for distributing a given total workload of size N on p processing elements such that the total computation time is minimized and resources are utilized efficiently. Results of our evaluation study show that the proposed method can speed up parallel applications significantly in compari- son to classical approaches. The proposed method is able to generate better solutions than classical methods in a reasonable amount of time by using a limited amount of prior information.

Suggestions

A Self-Cloning Agents Based Model for High-Performance Mobile-Cloud Computing
Angın, Pelin; Jin, Zhongjun (2015-07-02)
The rise of the mobile-cloud computing paradigm in recent years has enabled mobile devices with processing power and battery life limitations to achieve complex tasks in real-time. While mobile-cloud computing is promising to overcome the limitations of mobile devices for real-time computing, the lack of frameworks compatible with standard technologies and techniques for dynamic performance estimation and program component relocation makes it harder to adopt mobile-cloud computing at large. Most of the avai...
CLOUDGEN: Workload generation for the evaluation of cloud computing systems CLOUDGEN: Bulut Bilişim Sistemlerinin Başarim Deǧerlendirmesi icin Iş Yuku Uretimi
Koltuk, Furkan; Yazar, Alper; Schmidt, Şenan Ece (2019-04-01)
In this paper, we propose CLOUDGEN workflow that produces synthetic workloads for Infrastructure and Platform as a Service for the evaluation of resource management approaches in cloud computing systems. To this end, CLOUDGEN systematically processes and clusters records in a given workload trace and fits distributions for different workload parameters within the clusters. Different than the previous work, clustering is carried out to produce different virtual machine types for achieving models that are sui...
A reconfigurable computing platform for real time embedded applications
Say, Fatih; Halıcı, Uğur; Department of Electrical and Electronics Engineering (2011)
Today’s reconfigurable devices successfully combine ‘reconfigurable computing machine’ paradigm and ‘high degree of parallelism’ and hence reconfigurable computing emerged as a promising alternative for computing-intensive applications. Despite its superior performance and lower power consumption compared to general purpose computing using microprocessors, reconfigurable computing comes with a cost of design complexity. This thesis aims to reduce this complexity by providing a flexible and user friendly dev...
GPU algorithms for Efficient Exascale Discretizations
Abdelfattah, Ahmad; et. al. (2021-12-01)
In this paper we describe the research and development activities in the Center for Efficient Exascale Discretization within the US Exascale Computing Project, targeting state-of-the-art high-order finite-element algorithms for high-order applications on GPU-accelerated platforms. We discuss the GPU developments in several components of the CEED software stack, including the libCEED, MAGMA, MFEM, libParanumal, and Nek projects. We report performance and capability improvements in several CEED-enabled applic...
A Distributed Monitoring and Reconfiguration Approach for Adaptive Network Computing
Bhargava, Bharat; Angın, Pelin; Ranchal, Rohit; Lingayat, Sunil (2015-01-01)
The past decade has witnessed immense developments in the field of network computing thanks to the rise of the cloud computing paradigm, which enables shared access to a wealth of computing and storage resources without needing to own them. While cloud computing facilitates on-demand deployment, mobility and collaboration of services, mechanisms for enforcing security and performance constraints when accessing cloud services are still at an immature state. The highly dynamic nature of networks and clouds ma...
Citation Formats
M. Alasmar, “FPM Based Partitioning and Assignment Algorithm for Data Parallel Applications on Heterogeneous Platforms,” M.S. - Master of Science, Middle East Technical University, 2022.