MaxMem: Colocation and Performance for Big Data Applications on Tiered Main Memory Servers

arXiv - CS - Operating Systems Pub Date : 2023-12-01 DOI:arxiv-2312.00647

Amanda RaybuckThe University of Texas at Austin, Wei ZhangMicrosoft, Kayvan MansoorshahiThe University of Texas at Austin, Aditya K. KamathUniversity of Washington, Mattan ErezThe University of Texas at Austin, Simon PeterUniversity of Washington

{"title":"MaxMem: Colocation and Performance for Big Data Applications on Tiered Main Memory Servers","authors":"Amanda RaybuckThe University of Texas at Austin, Wei ZhangMicrosoft, Kayvan MansoorshahiThe University of Texas at Austin, Aditya K. KamathUniversity of Washington, Mattan ErezThe University of Texas at Austin, Simon PeterUniversity of Washington","doi":"arxiv-2312.00647","DOIUrl":null,"url":null,"abstract":"We present MaxMem, a tiered main memory management system that aims to\nmaximize Big Data application colocation and performance. MaxMem uses an\napplication-agnostic and lightweight memory occupancy control mechanism based\non fast memory miss ratios to provide application QoS under increasing\ncolocation. By relying on memory access sampling and binning to quickly\nidentify per-process memory heat gradients, MaxMem maximizes performance for\nmany applications sharing tiered main memory simultaneously. MaxMem is designed\nas a user-space memory manager to be easily modifiable and extensible, without\ncomplex kernel code development. On a system with tiered main memory consisting\nof DRAM and Intel Optane persistent memory modules, our evaluation confirms\nthat MaxMem provides 11% and 38% better throughput and up to 80% and an order\nof magnitude lower 99th percentile latency than HeMem and Linux AutoNUMA,\nrespectively, with a Big Data key-value store in dynamic colocation scenarios.","PeriodicalId":501333,"journal":{"name":"arXiv - CS - Operating Systems","volume":"19 1","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2023-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - CS - Operating Systems","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2312.00647","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}

引用次数: 0

Abstract

We present MaxMem, a tiered main memory management system that aims to maximize Big Data application colocation and performance. MaxMem uses an application-agnostic and lightweight memory occupancy control mechanism based on fast memory miss ratios to provide application QoS under increasing colocation. By relying on memory access sampling and binning to quickly identify per-process memory heat gradients, MaxMem maximizes performance for many applications sharing tiered main memory simultaneously. MaxMem is designed as a user-space memory manager to be easily modifiable and extensible, without complex kernel code development. On a system with tiered main memory consisting of DRAM and Intel Optane persistent memory modules, our evaluation confirms that MaxMem provides 11% and 38% better throughput and up to 80% and an order of magnitude lower 99th percentile latency than HeMem and Linux AutoNUMA, respectively, with a Big Data key-value store in dynamic colocation scenarios.

查看原文

微信好友朋友圈 QQ好友复制链接

本刊更多论文

MaxMem:分级主存服务器上大数据应用的托管和性能

我们介绍了MaxMem，一个分层主存管理系统，旨在最大限度地提高大数据应用的托管和性能。MaxMem使用基于快速内存缺失率的应用程序无关的轻量级内存占用控制机制，在不断增加的主机配置下提供应用程序QoS。通过依赖内存访问采样和分组来快速识别每个进程的内存热梯度，MaxMem最大限度地提高了同时共享分层主存的许多应用程序的性能。MaxMem被设计为一个用户空间内存管理器，可以很容易地修改和扩展，不需要复杂的内核代码开发。在一个由DRAM和英特尔Optane持久内存模块组成的分层主内存系统上，我们的评估证实，MaxMem在动态托管场景下，与HeMem和Linux AutoNUMA相比，MaxMem的吞吐量分别提高了11%和38%，延迟率高达80%，延迟率降低了99%。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文去求助

来源期刊

arXiv - CS - Operating Systems

自引率

0.00%

发文量

期刊最新文献

Analysis of Synchronization Mechanisms in Operating Systems Skip TLB flushes for reused pages within mmap's eBPF-mm: Userspace-guided memory management in Linux with eBPF BULKHEAD: Secure, Scalable, and Efficient Kernel Compartmentalization with PKS Rethinking Programmed I/O for Fast Devices, Cheap Cores, and Coherent Interconnects