Dynamic load balancing on heterogeneous multicore/multiGPU systems

Acosta, Alejandro; Corujo, Robert; Blanco, Vicente; Almeida, Francisco

doi:10.1109/hpcs.2010.5547097

Cited by 13 publications

(10 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…There are a lot of membership functions available, but this research chose the sigmoid function due to its "S" shaped curve can gracefully reflect the fluctuations of computational performance of GPU nodes [12]. According to equation (8), for instance, when a GPU's UF=0.6 at some point, then the membership value of UFL is 0.18, and the membership value of UFH is 0.82, see Figure 4.…”

Section: Fnn For Dynamic Load Balancingmentioning

confidence: 99%

“…Chen et al [7] proposed a task-based dynamic load balancing solution for multi-GPU systems that can achieve a near-linear speedup with the increase number of GPU nodes. Acosta et al [8] had developed a dynamic load balancing functional library that aims to balancing the load on each node according to the corresponding system runtime. However, these pilot studies are based on the assumptions that all GPU nodes equipped in a multi-GPU platform have equal computational capacity.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Dynamic load balancing on multi-GPUs system for big data processing

Zhang

Zhou

et al. 2017

2017 23rd International Conference on Automation and Computing (ICAC)

View full text Add to dashboard Cite

Abstract:The powerful parallel computing capability of modern GPU (Graphics Processing Unit) processors has attracted increasing attentions of researchers and engineers who had conducted a large number of GPU-based acceleration research projects. However, current single GPU based solutions are still incapable of fulfilling the real-time computational requirements from the latest big data applications. Thus, the multi-GPU solution has become a trend for many real-time application attempts. In those cases, the computational load balancing over the multiple GPU nodes is often the key bottleneck that needs to be further studied to ensure the best possible performance. The existing load balancing approaches are mainly based on the assumption that all GPUs in the same system provide equal computational performance, and had fallen short to address the situations from heterogeneous multi-GPU systems. This paper presents a novel dynamic load balancing model for heterogeneous multi-GPU systems based on the fuzzy neural network (FNN) framework. The devised model has been implemented and demonstrated in a case study for improving the computational performance of a two dimensional (2D) discrete wavelet transform (DWT). Experiment results show that this dynamic load balancing model has enabled a high computational throughput that can satisfy the real-time and accuracy requirements from many big data processing applications.

show abstract

Section: Fnn For Dynamic Load Balancingmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Dynamic load balancing on multi-GPUs system for big data processing

Zhang

Zhou

et al. 2017

2017 23rd International Conference on Automation and Computing (ICAC)

View full text Add to dashboard Cite

show abstract

“…Previous work has described effective approaches for load-balancing applications with multiple concurrently executing kernels [9,12,32,38,98,103] or applications in which a single kernel is run repeatedly and consumes its own output [1,26]. The challenge in the former case is determining on which device to execute the entirety of each kernel, while the challenge in the latter case is refining the work partition in each iteration while taking into account data locality.…”

Section: Chapter 4 Heterogeneous Load Balancingmentioning

confidence: 99%

“…However, even 1 We define efficiency as the measured throughput divided by the sum of the throughputs of the devices when executing separately. with relatively large differences in performance, load balancing can still provide non-trivial speedups.…”

Section: Need For Heterogeneous Load Balancingmentioning

confidence: 99%

See 1 more Smart Citation

Improving Resource Utilization in Heterogeneous CPU-GPU Systems

Boyer¹

View full text Add to dashboard Cite

Graphics processing units (GPUs) have attracted enormous interest over the past decade due to substantial increases in both performance and programmability. Programmers can potentially leverage GPUs for substantial performance gains, but at the cost of significant software engineering effort. In practice, most GPU applications do not effectively utilize all of the available resources in a system: they either fail to use use a resource at all or use a resource to less than its full potential. This underutilization can hurt both performance and energy efficiency. In this dissertation, we address the underutilization of resources in heterogeneous CPU-GPU systems in three different contexts.First, we address the underutilization of a single GPU by reducing CPU-GPU interaction to improve performance. We use as a case study a computationally-intensive video-tracking application from systems biology. Because of the high cost of CPU-GPU coordination, our initial, straightforward attempts to accelerate this application failed to effectively utilize the GPU. By leveraging some non-obvious optimization strategies, we significantly decreased the amount of CPU-GPU interaction and improved the performance of the GPU implementation by 26x relative to the best CPU implementation. Based on the lessons we learned, we present general guidelines for optimizing GPU applications as well as recommendations for system-level changes that would simplify the development of high-performance GPU applications.Next, we address underutilization at the system level by using load balancing to improve performance. We propose a dynamic scheduling algorithm that automatically and efficiently i Abstract ii divides the execution of a data-parallel kernel across multiple, possibly heterogeneous GPUs.We show that our scheduler can nearly match the performance of an unrealistic static scheduler when device performance is fixed, and can provide better performance when device performance varies.Finally, we address underutilization within a GPU by using frequency scaling to improve energy efficiency. We propose a novel algorithm for predicting the energy-optimal GPU clock frequencies for an arbitrary kernel. Using power measurements from real systems, we demonstrate that our algorithm improves significantly on the state of the art across multiple generations of GPUs. We also propose and evaluate techniques for decreasing the CPU's energy consumption during GPU computation.Many of the techniques presented in this dissertation can be used to improve the performance and energy efficiency of GPU applications with no programmer effort or software modifications required. As the diversity of available hardware systems continues to increase, automatic techniques such as these will become critical for software to fully realize the benefits of future hardware improvements.Approval Sheet

show abstract

Parallel solution of the subset‐sum problem: an empirical study

Bokhari

2012

Concurrency and Computation

View full text Add to dashboard Cite

The subset-sum problem is a well-known NP-complete combinatorial problem that is solvable in pseudopolynomial time, that is, time proportional to the number of input objects multiplied by the sum of their sizes. This product defines the size of the dynamic programming table used to solve the problem. We show how this problem can be parallelized on three contemporary architectures, that is, a 128-processor Cray Extreme Multithreading (XMT) massively multithreaded machine, a 16-processor IBM x3755 shared memory machine, and a 240-core NVIDIA FX 5800 graphics processing unit (GPU). We show that it is straightforward to parallelize this algorithm on the Cray XMT primarily because of the word-level locking that is available on this architecture. For the other two machines, we present an alternating word algorithm that can implement an efficient solution. Our results show that the GPU performs well for problems whose tables fit within the device memory. Because GPUs typically have memories in the order of 10 GB, such architectures are best for small problem sizes that have tables of size approximately 10 10 . The IBM x3755 performs very well on medium-sized problems that fit within its 64-GB memory but has poor scalability as the number of processors increases and is unable to sustain performance as the problem size increases. This machine tends to saturate for problem sizes of 10 11 bits. The Cray XMT shows very good scaling for large problems and demonstrates sustained performance as the problem size increases. However, this machine has poor scaling for small problem sizes; it performs best for problem sizes of 10 12 bits or more. The results in this paper illustrate that the subset-sum problem can be parallelized well on all three architectures, albeit for different ranges of problem sizes. The performance of these three machines under varying problem sizes show the strengths and weaknesses of the three architectures.

show abstract

Dynamic load balancing on heterogeneous multicore/multiGPU systems

Cited by 13 publications

References 7 publications

Dynamic load balancing on multi-GPUs system for big data processing

Dynamic load balancing on multi-GPUs system for big data processing

Improving Resource Utilization in Heterogeneous CPU-GPU Systems

Parallel solution of the subset‐sum problem: an empirical study

Contact Info

Product

Resources

About