2011 International Conference on Parallel Processing 2011
DOI: 10.1109/icpp.2011.88
|View full text |Cite
|
Sign up to set email alerts
|

GPU Resource Sharing and Virtualization on High Performance Computing Systems

Abstract: Modern Graphic Processing Units (GPUs) are widely used as application accelerators in the High Performance Computing (HPC) field due to their massive floating-point computational capabilities and highly dataparallel computing architecture. Contemporary high performance computers equipped with co-processors such as GPUs primarily execute parallel applications using the Single Program Multiple Data (SPMD) model, which requires balanced computing resources of both microprocessor and coprocessors to ensure full sy… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
1
1
1

Citation Types

0
26
0

Year Published

2013
2013
2018
2018

Publication Types

Select...
5
2

Relationship

1
6

Authors

Journals

citations
Cited by 43 publications
(26 citation statements)
references
References 14 publications
0
26
0
Order By: Relevance
“…The other category of solutions is to provide process-level GPU sharing. Our previous work [18] presented a GPU virtualization infrastructure that provides a virtual SPMD model by exposing multiple virtual GPU interfaces to the processors. The virtualization infrastructure allows for multiple processes to share the GPU using a single GPU context and to concurrently execute GPU kernels, as well as to achieve concurrency between data transfer and kernel execution.…”
Section: Related Workmentioning
confidence: 99%
See 2 more Smart Citations
“…The other category of solutions is to provide process-level GPU sharing. Our previous work [18] presented a GPU virtualization infrastructure that provides a virtual SPMD model by exposing multiple virtual GPU interfaces to the processors. The virtualization infrastructure allows for multiple processes to share the GPU using a single GPU context and to concurrently execute GPU kernels, as well as to achieve concurrency between data transfer and kernel execution.…”
Section: Related Workmentioning
confidence: 99%
“…As we explained earlier, the latest Kepler GPU architecture [19] provides native hardware Hyper-Q support, which allows multiple processes to share a single GPU context using the CUDA proxy server feature. As the purposes of utilizing our GPU virtualization approach (for Fermi or earlier GPUs), as well as using the Hyper-Q feature (for Kepler series of GPUs) are both to meet the single GPU context requirement for efficient GPU sharing, here, we provide a brief description of the GPU virtualization approach addressed by our previous work [18]. Figure 2 shows that all SPMD GPU kernels are executed within the single daemon process using CUDA streams.…”
Section: Gpu Sharing Approach With Streams For Spmd Programsmentioning
confidence: 99%
See 1 more Smart Citation
“…Li et al [20] introduced a virtualization layer that makes all the participating processes execute kernels in the same GPU context, similar to NVIDIA MPS [24]. GERM [7] and TimeGraph [17] focus on graphics applications and provide a GPU command schedulers integrated in the device driver.…”
Section: Related Workmentioning
confidence: 99%
“…To overcome the inefficiencies introduced by multiple processes sharing the GPU [20], NVIDIA provides a software solution called Muli-Process Service (MPS) [24]. MPS instantiates a proxy process that receives requests from client processes (e.g., processes in an MPI application) and executes them on the GPU.…”
Section: Gpu Program Executionmentioning
confidence: 99%