Towards the Dynamic Load Balancing on Heterogeneous Multi-GPU Systems

Acosta, Alejandro; Blanco, Vicente; Almeida, Francisco

doi:10.1109/ispa.2012.96

Cited by 18 publications

(14 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Most powerful platforms for a wide class of Many parallel programs embrace the concept of many-core processors, which can be accomplished via Compiler based, Application Programming Interface (API) such as OpenMP parallel programming. In the other hand multi-GPUs can be interfaced to on-chip multicores as efficient accelerators to increase system-level computational performance and improve system responsiveness [1,2]. For instance, the NVIDIA Tesla S1070 GPU [20] consists of (30×4) multiprocessors, where each multiprocessor contains 8 cores, resulting in an aggregate number of 960 cores inside a single GPU.…”

Section: Introductionmentioning

confidence: 99%

Global Hybrid Multi-core-GPUs-OpenMPs-Resources Platform in Hard Real Time System

Rabee

Liao

Yang

et al. 2014

2014 IEEE 17th International Conference on Computational Science and Engineering

View full text Add to dashboard Cite

Interesting with technology of parallel processing and multi-core system, we present a new design for multi-core-GPUs platform system with different shared resources, as well as get the advantage of OpenMP parallel programming to design a new unit for real time system where this platform divided into two part; first: clustering the multi-core-GPUs-OpenMP unit -shared resources by a new method, where CPUs and GPUs grouped in one cluster with OpenMP unit and grouped the shared resources by unique global cluster and scheduled it by global scheduling as well as the local locking shared resources protocol. Second; present an innovative mechanism for scheduling GPUs-OpenMP unit with shared resources in multi-core system. The GPU has been tested with many operations by OpenCL language and used the GPU performance with multi-core system depending on local and distributed semaphores evaluations and compare it with MPCP protocol. The evaluations of our proposal were very good.

show abstract

Section: Introductionmentioning

confidence: 99%

Global Hybrid Multi-core-GPUs-OpenMPs-Resources Platform in Hard Real Time System

Rabee

Liao

Yang

et al. 2014

2014 IEEE 17th International Conference on Computational Science and Engineering

View full text Add to dashboard Cite

show abstract

“…Most powerful platforms for a wide class of many parallel programs embrace the concept of manycore processors, which can be accomplished via Compiler based, Application. Multi-GPUs can be interfaced to on-chip multicores as efficient accelerators to increase system-level computational performance and improve system responsiveness [1,2]. For instance, the NVIDIA Tesla S1070 GPU [20] consists of (30×4) multiprocessors, where each multiprocessor contains 8 cores, resulting in an aggregate number of 960 cores inside a single GPU.…”

Section: Introductionmentioning

confidence: 99%

Global hybrid multi-core-GPUs-resources platform (GHRP) in hard real time system

Rabee

Liao

Yang

et al. 2014

2014 7th International Conference on Biomedical Engineering and Informatics

View full text Add to dashboard Cite

Graphics Processing Units (GPU) technologies have been adopted in many real-time applications, due to the developments in the architectures of GPU fit to be used in general purpose operation. Interesting with that we present a platform for hybrid multi-core-GPUs with global/local shared resources, where this platform divided into three parts; first, cluster the multi-core-GPUs-shared resources, where CPUs and GPUs grouped in one cluster with local shared resources and grouped the global shared resources by unique global cluster where the shared resources governed by global and local semaphore. Second, present an innovative mechanism for scheduling GPUs with different shared resources in multi-core system. Third, drive a precise blocking time. The GPU has been tested with many operations by OpenCL language and used the GPU performance with multi-core system depending on local and distributed semaphores evaluations and compare it with MPCP protocol.

show abstract

“…They hid the other details of the possible differences between processors. In [1], Acosta et al worked in CPU-GPU. They also measured the processing time.…”

Section: Introductionmentioning

confidence: 99%

“…It is not clear whether this technique is employed in [1] and [7], but it is used in all the other ones mentioned above. Another technique is asymmetry load balancing, which means tasks are better distributed to the processors according to their capabilities, instead of distributing to them evenly.…”

Section: Introductionmentioning

confidence: 99%

“…Another technique is asymmetry load balancing, which means tasks are better distributed to the processors according to their capabilities, instead of distributing to them evenly. All the above papers apply this technique, except that it is unclear in [1]. To the best of our knowledge, there is no practical work on bicriteria load balancing on heterogeneous systems.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Bicriteria load balancing for online placement in heterogeneous servers with totally-ordered upper bounds

Tse

2013

2013 Second International Conference on Informatics &Amp; Applications (ICIA)

View full text Add to dashboard Cite

We study the online load balancing problem for two independent criteria in heterogeneous systems. For convenience, we choose a system of distributed file servers located in a cluster as the scenario, although our work is not limited to it. Every server is assigned upper bounds for its load and storage space. We assume that the heterogeneity of servers is eventually reflected by the difference of these assignments among different servers. In this paper, the upper bounds are totally ordered (in the sense that, for any two servers, one must be more capable than another). We propose a solution for balancing their loads and required storage spaces for online document placements, and document reallocation is allowed. The time complexity is (log ), plus the reallocation cost incurred, where is the number of servers, and is the number of existing documents in the system. This cost is a measure of the price for coping with the heterogeneity of our model.

show abstract

Towards the Dynamic Load Balancing on Heterogeneous Multi-GPU Systems

Cited by 18 publications

References 11 publications

Global Hybrid Multi-core-GPUs-OpenMPs-Resources Platform in Hard Real Time System

Global Hybrid Multi-core-GPUs-OpenMPs-Resources Platform in Hard Real Time System

Global hybrid multi-core-GPUs-resources platform (GHRP) in hard real time system

Bicriteria load balancing for online placement in heterogeneous servers with totally-ordered upper bounds

Contact Info

Product

Resources

About