Name
Abstract | ||
|---|---|---|
Lately, cloud computing has seen explosive growth, due to the flexibility and scalability it offers. The ever-increasing computational demands, especially from the machine learning domain, have forced cloud operators to enhance their infrastructure with acceleration devices, such as General-Purpose (GP)GPUs or FPGAs. Even though multi-tenancy has been widely examined for conventional CPUs, this is not the case for accelerators. Current solutions support "one accelerator per user" schemes, which can lead to both under-utilization and starvation of available resources.
In this work, we analyze the potentials of GPU sharing inside data-center environments. We investigate how several architectural features affect the performance of GPUs under different multi-tenant stressing scenarios. We compare CUDA MPS with the native, default CUDA scheduler and also with Vinetalk, a research framework providing GPU sharing capabilities. Experimental results show that NVIDIA's MPS achieves the best performance in multi-application scenarios, specifically up to X4.5 and X11.2 compared to native CUDA scheduler and Vinetalk respectively.
|
Year | DOI | Venue |
|---|---|---|
2020 | 10.1145/3378678.3391887 | SCOPES '20: 23rd International Workshop on Software and Compilers for Embedded Systems
St. Goar
Germany
May, 2020 |
Keywords | DocType | ISBN |
GPGPU sharing, native CUDA queues, CUDA MPS, Vinetalk, interference analysis, cloud computing | Conference | 978-1-4503-7131-5 |
Citations | PageRank | References |
1 | 0.37 | 0 |
Authors | ||
5 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| Ioannis Oroutzoglou | 1 | 3 | 0.75 |
| Dimosthenis Masouros | 2 | 12 | 6.37 |
| Konstantina Koliogeorgi | 3 | 1 | 2.06 |
| Sotirios Xydis | 4 | 144 | 31.51 |
| Dimitrios Soudris | 5 | 243 | 48.41 |