AsmDB: Understanding and Mitigating Front-End Stalls in Warehouse-Scale Computers

被引：4

作者：

Nagendra, Nayana Prasad ^{[1
]}

Ayers, Grant ^{[2
]}

August, David I. ^{[3
]}

Cho, Hyoun Kyu ^{[2
]}

Kanev, Svilen ^{[4
]}

Kozyrakis, Christos ^{[5
]}

Krishnamurthy, Trivikram ^{[6
]}

Litz, Heiner ^{[7
]}

Moseley, Tipp ^{[8
]}

Ranganathan, Parthasarathy ^{[9
]}

机构：

[1] Princeton Univ, Dept Comp Sci, Princeton, NJ 08544 USA

[2] Google, Mountain View, CA USA

[3] Princeton Univ, Dept Comp Sci, Liberty Res Grp, Princeton, NJ 08544 USA

[4] Google, Translating Datactr Performance Anal Insights Per, Mountain View, CA USA

[5] Stanford Univ, Elect Engn & Comp Sci, Stanford, CA 94305 USA

[6] Nvidia, Santa Clara, CA USA

[7] Univ Calif Santa Cruz, Comp Sci & Engn Dept, Santa Cruz, CA 95064 USA

[8] Google, Datactr Scale Performance Anal, Mountain View, CA USA

[9] Google, Nextgenerat Syst, Mountain View, CA USA

来源：

IEEE MICRO | 2020年 / 40卷 / 03期

关键词：

Prefetching; Optimization; Servers; Hardware; Databases; Complexity theory;

D O I：

10.1109/MM.2020.2986212

中图分类号：

TP3 [计算技术、计算机技术];

学科分类号：

0812 ;

摘要：

It is well known that the datacenters hosting today's cloud services waste a significant number of cycles on front-end stalls. However, prior work has provided little insights about the source of these front-end stalls and how to address them. This work analyzes the cause of instruction cache misses at a fleet-wide scale and proposes a new compiler-driven software code prefetching strategy to reduce instruction caches misses by 90%.

引用

下载

页码：56 / 63

页数：8