paperbot · PL 论文追踪

全部论文 待读 收藏

P2 AvA: Accelerated Virtualization of Accelerators

AvA:加速器的加速虚拟化
ASPLOS Proceedings of the Twenty-Fifth International Conference on Architectural Support for Programming Languages and Operating Systems2020匹配度 1/10 · 该论文研究 GPU/TPU/FPGA 等加速器的虚拟化与多租户共享,属于云系统/运行时方向,不涉及静态分析、形式化验证、浮点程序分析、贝叶斯程序分析或 LLM/Agent 与程序分析的交叉。虽然与机器学习硬件系统有微弱关联,但不在用户研究画像的核心或邻近范围内。引用 33 inbox
Hangchen Yu, Arthur Michener Peters, Amogh Akshintala, Christopher J. Rossbach
crossref: 2026-09-12

AI 速览

一句话:提出 AvA,为 GPU/TPU/FPGA 等加速器提供加速虚拟化,使云中多个 guest 可共享物理设备,恢复多租户与整合。

云计算应用大规模迁移到云端,而 GPU、TPU、FPGA 等加速器在后摩尔定律时代不断普及。二者存在冲突:云应用运行在虚拟化平台上,但现有虚拟化技术尚未为加速器提供生产可用的方案。云提供商因此通常将物理加速器设备独占分配给单个 guest,导致多租户与资源整合能力丧失。AvA 提出面向加速器的加速虚拟化(accelerated virtualization of accelerators),旨在让多个 guest 共享加速器,从而在云环境中恢复多租户和 consolidation。摘要未披露具体实现机制与实验评估结果。

问题
云应用运行在虚拟平台上,但现有技术无法为 GPU/TPU/FPGA 等加速器提供生产级虚拟化,迫使云提供商将物理设备独占给单个 guest,牺牲多租户与整合。
方法
提出 AvA,从标题与摘要看,其核心是为加速器提供加速虚拟化,使多个 guest 能共享加速器。
评估
摘要未给出实验评估信息;仅说明该工作旨在恢复云中加速器的多租户与 consolidation。
原文摘要(Abstract)

Applications are migrating en masse to the cloud, while accelerators such as GPUs, TPUs, and FPGAs proliferate in the wake of Moore's Law. These trends are in conflict: cloud applications run on virtual platforms, but existing virtualization techniques have not provided production-ready solutions for accelerators. As a result, cloud providers expose accelerators by dedicating physical devices to individual guests. Multi-tenancy and consolidation are lost as a consequence.

链接与引用

DOI 原文 · PDF(开放获取) ·

BibTeX
@inproceedings{paperbot5416,
  title = {AvA: Accelerated Virtualization of Accelerators},
  author = {Hangchen Yu and Arthur Michener Peters and Amogh Akshintala and Christopher J. Rossbach},
  booktitle = {ASPLOS},
  year = {2020},
  doi = {10.1145/3373376.3378466}
}