Early Experiences in Running Many-Task Computing Workloads on GPGPUs
نویسندگان
چکیده
This work aims to enable Swift to efficiently use accelerators (such as NVIDIA GPUs) to further accelerate a wide range of applications. This work presents preliminary results in the costs associated with managing and launching concurrent kernels on NVIDIA Kepler GPUs. We expect our results to be applicable to several XSEDE resources, such as Forge, Keeneland, and Lonestar, where currently Swift can only use the general processors to execute workloads and the GPUs are left idle. Keywords-Many-Task Computing, Swift, GPGPU, CUDA
منابع مشابه
My Cray can do that? Supporting Diverse Workloads on the Cray XE-6
The Cray XE architecture has been optimized to support tightly coupled MPI applications, but there is an increasing need to run more diverse workloads in the scientific and technical computing domains. These needs are being driven by trends such as the increasing need to process “Big Data”. In the scientific arena, this is exemplified by the need to analyze data from instruments ranging from se...
متن کاملAn Efficient Genetic Algorithm for Task Scheduling on Heterogeneous Computing Systems Based on TRIZ
An efficient assignment and scheduling of tasks is one of the key elements in effective utilization of heterogeneous multiprocessor systems. The task scheduling problem has been proven to be NP-hard is the reason why we used meta-heuristic methods for finding a suboptimal schedule. In this paper we proposed a new approach using TRIZ (specially 40 inventive principles). The basic idea of thi...
متن کاملAn Efficient Genetic Algorithm for Task Scheduling on Heterogeneous Computing Systems Based on TRIZ
An efficient assignment and scheduling of tasks is one of the key elements in effective utilization of heterogeneous multiprocessor systems. The task scheduling problem has been proven to be NP-hard is the reason why we used meta-heuristic methods for finding a suboptimal schedule. In this paper we proposed a new approach using TRIZ (specially 40 inventive principles). The basic idea of thi...
متن کاملTowards Next Generation Resource Management at Extreme-Scales
With the exponential growth of distributed systems in both FLOPS and parallelism (number of cores/threads), scientific applications are growing more diverse with various workloads. These workloads include traditional large-scale high performance computing (HPC) MPI jobs, and HPC ensemble workloads that support the investigation of parameter sweeps using many small-scale coordinated jobs, as wel...
متن کاملChallenge Benchmarks That Must be Conquered to Sustain the GPU Revolution
The shift from GPUs to GPGPUs has brought with it many changes to the GPU architecture (e.g. more caches, more concurrent kernels, better synchronization). As GPUs press further into the general-purpose domain, architects must continue to address the performance of challenging workloads. This paper presents a set of challenge benchmarks and their key performance limitations to help direct futur...
متن کامل