paperbot · PL 论文追踪

全部论文 待读 收藏

P2 LATR

LATR:惰性 TLB 一致性机制
ASPLOS Proceedings of the Twenty-Third International Conference on Architectural Support for Programming Languages and Operating Systems2018匹配度 0/10 · 该论文研究操作系统层面的惰性 TLB 一致性机制,属于运行时与内存管理方向,与用户关注的静态分析、LLM/Agent 程序分析、浮点程序分析及贝叶斯程序分析均无交集。引用 34 inbox
Mohan Kumar Kumar, Steffen Maass, Sanidhya Kashyap, Ján Veselý, Zi Yan, Taesoo Kim, Abhishek Bhattacharjee, Tushar Krishna
crossref: 2026-09-12

AI 速览

一句话:提出软件实现的惰性 TLB 一致性机制,规避跨核 IPI 开销,显著加速虚拟内存操作。

现有操作系统采用同步 TLB shootdown 维持 TLB 一致性,需要向远端核发送 IPI 并执行中断处理,给 free、页迁移等虚拟内存操作带来显著开销。LATR 提出一种基于软件的惰性 TLB 一致性机制,以惰性方式处理 TLB 失效,从而避免昂贵的 IPI 投递及相关中断处理开销。评估表明,在 2-socket 机器上 LATR 将 munmap() 延迟降低 70.8%;无需修改应用,Apache 相比 Linux 提升 59.9%,相比高度优化的 ABIS 提升 37.9%。

问题
现有操作系统的同步 TLB shootdown 需通过 IPI 向远端核发送失效信号并执行中断处理,导致 free、页迁移等虚拟内存操作开销高昂。
方法
LATR 以软件方式惰性处理 TLB 一致性,避免为传递 shootdown 信号而使用昂贵的 IPI 及关联中断处理开销。
评估
在 2-socket 机器上评估,LATR 将 munmap() 延迟降低 70.8%;无需应用改动,Apache 相比 Linux 提升 59.9%,相比 ABIS 提升 37.9%。
原文摘要(Abstract)

We propose LATR-lazy TLB coherence-a software-based TLB shootdown mechanism that can alleviate the overhead of the synchronous TLB shootdown mechanism in existing operating systems. By handling the TLB coherence in a lazy fashion, LATR can avoid expensive IPIs which are required for delivering a shootdown signal to remote cores, and the performance overhead of associated interrupt handlers. Therefore, virtual memory operations, such as free and page migration operations, can benefit significantly from LATR's mechanism. For example, LATR improves the latency of munmap() by 70.8% on a 2-socket machine, a widely used configuration in modern data centers. Real-world, performance-critical applications such as web servers can also benefit from LATR: without any application-level changes, LATR improves Apache by 59.9% compared to Linux, and by 37.9% compared to ABIS, a highly optimized, state-of-the-art TLB coherence technique.

链接与引用

DOI 原文 ·

BibTeX
@inproceedings{paperbot4255,
  title = {LATR},
  author = {Mohan Kumar Kumar and Steffen Maass and Sanidhya Kashyap and Ján Veselý and Zi Yan and Taesoo Kim and Abhishek Bhattacharjee and Tushar Krishna},
  booktitle = {ASPLOS},
  year = {2018},
  doi = {10.1145/3173162.3173198}
}