.png){kind=icon}
.png){kind=icon}
Return
Efficient accelerator sharing in virtualized environments: A Xeon Phi use-case
.png){kind=icon}
PRE
AI
DOI：10.1016/j.jss.2018.12.029.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
High performance computing experts and scientists increasingly pay attention to heterogeneous processing and most systems today are built with some form of heterogeneity. At the same time, a large amount of HPC workloads is executed in cloud computing environments, due to the benefits that cloud offers both for the users and the infrastructure providers. In this context, there is a growing need to bridge the gap between the accelerator ecosystem and the world of virtualization, which is the building block of cloud computing environments. Hence, future accelerator platforms need to be designed in a virtualization-friendly approach, both in the software and in the hardware level. In this paper we extend our previous work and introduce vPHI, a transparent framework that enables sharing of Xeon Phi devices between multiple virtual machines. We design vPHI based on the paravirtualization technique and provide compatibility with the accelerator's native transport layer to the upper level software stack components. In this way, with vPHI existing applications can be used without any source code or binary modifications. The evaluation of our framework in the popular offload mode of execution shows promising results with vPHI exhibiting minimal overhead for the computational phases in a deployment with multiple (up to 8) VMs. vPHI can enable better accelerator utilization, increasing up to 3.4x the total throughput versus a single host application. We also discuss the applicability of our approach to future accelerator technologies and in this way form a path to bring the accelerator use cases closer to the cloud computing world. (C) 2019 Elsevier Inc. All rights reserved.
Keywords:
vPHI
Accelerator virtualization
Accelerator sharing
Virtualization
HPC
Xeon Phi virtualization
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
4.1
Papers:
5.4K
Citations:
8.4K