Adaptive Intelligent Tiering for modern storage systems

IF 1 4区计算机科学 Q4 COMPUTER SCIENCE, HARDWARE & ARCHITECTURE

Performance Evaluation Pub Date : 2023-05-01 DOI:10.1016/j.peva.2023.102332

Lu Pang , Anis Alazzawe , Madhurima Ray , Krishna Kant , Jeremy Swift

{"title":"Adaptive Intelligent Tiering for modern storage systems","authors":"Lu Pang , Anis Alazzawe , Madhurima Ray , Krishna Kant , Jeremy Swift","doi":"10.1016/j.peva.2023.102332","DOIUrl":null,"url":null,"abstract":"<div><p><span>Enterprise systems routinely use tiered storage consisting of a hierarchy of storage devices that vary in speed and size. One key to obtaining good performance in such a hierarchy is to migrate data elements intelligently to the appropriate tier. For example, moving the most used data towards the fastest tier and the least used data towards the slowest tier. Tiering is typically done based on usage </span>statistics<span><span> over relatively long time periods. In this paper, we consider a much more agile tiering mechanism called Adaptive Intelligent Tiering (AIT). It can dynamically adapt to the changing behavior of storage accesses by the running applications. The AIT mechanism uses a deep learning model to generate a set of candidate movements and employs a </span>reinforcement learning mechanism to further refine the candidates. Based on extensive simulations in a 3-tier system, we show that the proposed scheme, compared with several other methods, enhances workload performance up to 85% on storage traces with a wide range of characteristics.</span></p></div>","PeriodicalId":19964,"journal":{"name":"Performance Evaluation","volume":"160 ","pages":"Article 102332"},"PeriodicalIF":1.0000,"publicationDate":"2023-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Performance Evaluation","FirstCategoryId":"94","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0166531623000020","RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q4","JCRName":"COMPUTER SCIENCE, HARDWARE & ARCHITECTURE","Score":null,"Total":0}

引用次数: 0

Abstract

Enterprise systems routinely use tiered storage consisting of a hierarchy of storage devices that vary in speed and size. One key to obtaining good performance in such a hierarchy is to migrate data elements intelligently to the appropriate tier. For example, moving the most used data towards the fastest tier and the least used data towards the slowest tier. Tiering is typically done based on usage statistics over relatively long time periods. In this paper, we consider a much more agile tiering mechanism called Adaptive Intelligent Tiering (AIT). It can dynamically adapt to the changing behavior of storage accesses by the running applications. The AIT mechanism uses a deep learning model to generate a set of candidate movements and employs a reinforcement learning mechanism to further refine the candidates. Based on extensive simulations in a 3-tier system, we show that the proposed scheme, compared with several other methods, enhances workload performance up to 85% on storage traces with a wide range of characteristics.

查看原文本刊更多论文

现代存储系统的自适应智能分层

企业系统通常使用分层存储，分层存储由不同速度和大小的存储设备组成。在这种层次结构中获得良好性能的一个关键是将数据元素智能地迁移到适当的层。例如，将使用最多的数据移动到最快的层，将使用最少的数据移动到最慢的层。分级通常基于相对较长时间内的使用统计数据。在本文中，我们考虑了一种更敏捷的分层机制，称为自适应智能分层(AIT)。它可以动态地适应运行中的应用程序对存储访问行为的变化。AIT机制使用深度学习模型生成一组候选运动，并使用强化学习机制进一步细化候选。基于在三层系统中的大量模拟，我们表明，与其他几种方法相比，所提出的方案在具有广泛特征的存储走线上将工作负载性能提高了85%。

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分钟内获得全文求助全文

来源期刊

Performance Evaluation 工程技术-计算机：理论方法

CiteScore

3.10

自引率

0.00%

发文量

审稿时长

24 days

期刊介绍： Performance Evaluation functions as a leading journal in the area of modeling, measurement, and evaluation of performance aspects of computing and communication systems. As such, it aims to present a balanced and complete view of the entire Performance Evaluation profession. Hence, the journal is interested in papers that focus on one or more of the following dimensions: -Define new performance evaluation tools, including measurement and monitoring tools as well as modeling and analytic techniques -Provide new insights into the performance of computing and communication systems -Introduce new application areas where performance evaluation tools can play an important role and creative new uses for performance evaluation tools. More specifically, common application areas of interest include the performance of: -Resource allocation and control methods and algorithms (e.g. routing and flow control in networks, bandwidth allocation, processor scheduling, memory management) -System architecture, design and implementation -Cognitive radio -VANETs -Social networks and media -Energy efficient ICT -Energy harvesting -Data centers -Data centric networks -System reliability -System tuning and capacity planning -Wireless and sensor networks -Autonomic and self-organizing systems -Embedded systems -Network science