优化RocksDB以提升Redis在闪存上的性能

需积分: 9 72 浏览量更新于2024-08-28 收藏 392KB PDF 举报

"这篇文档是关于在闪存上优化RocksDB以提升Redis性能的研究报告。作者们来自希伯来大学、RedisLabs和三星半导体，他们深入探讨了如何通过调整RocksDB的配置来最大化利用固态驱动器（SSD）的优势，特别是在Redis-on-Flash（RoF）系统中的应用。" 在现代IT环境中，RocksDB是一个广泛使用的键值存储系统，专为高速存储而优化。随着固态驱动器（SSDs）的普及，RocksDB在生产环境中的应用越来越普遍，尤其被用作存储引擎来加速对块存储的访问。然而，RocksDB的调优是一项复杂的任务，涉及众多参数且相互间有不同程度的依赖关系。这篇论文揭示了一个经过精细调优的配置可以将性能提升一个数量级，相对于默认配置而言，这是一个显著的改进。作者们专注于优化RocksDB以适应Redis-on-Flash（RoF）的场景，RoF是一个商业实现的Redis内存键值存储系统，它利用SSD作为RAM的扩展，极大地提高了单节点的有效容量。RoF将热值存储在快速的SSD上，以利用其高I/O速度，同时降低了内存成本。在优化过程中，作者们可能考虑了以下几个关键知识点： 1. **参数调优**：包括但不限于块缓存大小、写缓冲区大小、压缩选项、布隆过滤器设置等，每个参数都会对系统的读写性能、内存使用和I/O效率产生影响。 2. **工作负载分析**：理解RoF的工作负载特性，如读写比例、数据分布、访问模式等，对于选择合适的优化策略至关重要。 3. **SSD特性的利用**：充分利用SSD的低延迟和高吞吐量，例如通过优化写入策略来减少随机写入，增加顺序写入。 4. **并发控制**：优化多线程和多核心环境下的并发处理能力，以最大化硬件资源利用率。 5. **缓存策略**：调整缓存策略以优化热数据的访问，减少对慢速存储的访问。 6. **日志管理**：优化日志写入和回放过程，以减少延迟并提高恢复速度。 7. **压缩算法选择**：根据数据特性选择合适的压缩算法，平衡压缩效率与CPU使用率。 8. **故障恢复和数据持久化**：优化这些过程以降低对系统性能的影响。通过这样的深度优化，RoF能够提供更高的吞吐量、更低的延迟以及更稳定的性能，这对于需要处理大量数据的现代应用来说是非常关键的。这不仅有助于提升用户体验，还能降低运营成本，因为使用SSD作为RAM扩展可以减少对昂贵的DRAM的需求。这篇论文为在闪存上优化RocksDB提供了宝贵的经验和深入洞察，对于那些在大规模数据存储和处理环境中使用RocksDB和SSD的系统管理员和开发者来说，具有很高的参考价值。

Optimization of RocksDB for Redis on Flash

Keren Ouaknine

Hebrew University

Givat Ram Jerusalem

9190401 Israel

ouaknine@cs.huji.ac.il

Oran Agra

Redis Labs

Habarzel 28 Tel-Aviv

6971040 Israel

oran@redislabs.com

Zvika Guz

Samsung Semiconductor

3655 N 1st st. San Jose CA

95134 USA

zvika.guz@samsung.com

ABSTRACT

RocksDB is a popular key-value store, optimized for fast storage.

With Solid-State Drives (SSDs) becoming prevalent, RocksDB

gained widespread adoption and is now common in production set-

tings. Speciﬁcally, various software stacks embed RocksDB as a

storage engine to optimize access to block storage. Unfortunately,

tuning RocksDB is a complex task, involving many parameters

with different degrees of dependencies. As we show in this pa-

per, a highly tuned conﬁguration can improve performance by an

order of magnitude over the baseline conﬁguration.

In this paper, we describe our experience optimizing RocksDB for

Redis-on-Flash (RoF) – a commercial implementation of the Redis

in-memory key-value store that uses SSDs as RAM extension to

dramatically increase the effective per-node capacity. RoF stores

hot values in RAM, and utilizes RocksDB to store and manage

cold data on SSD drives. We describe our methodology for tun-

ing RocksDB parameters and present our experiments and ﬁnd-

ings (including both positive and negative tuning results) on two

clouds: EC2 and GCE. Overall, we show how tuning RocksDB im-

proved the database replication time for RoF by more than 11x. We

hope that our experience will help others adopt, conﬁgure, and tune

RocksDB in order to realize its full performance potential.

CCS Concepts

•Information systems → Key-value stores; Database perfor-

mance evaluation;

Keywords

Databases, Benchmark, Redis, Rocksdb, Key-Value Store, SSD,

NVMe

1. INTRODUCTION

RocksDB is a persistent key-value (KV) store that was speciﬁcally

architected for fast storage, mainly ﬂash-based SSDs [1]. Forked

from LevelDB [2], RocksDB provides superior performance [3],

and was designed to be highly ﬂexible in order to facilitate embed-

ding as a storage engine by higher-level applications. Indeed, many

Permission to make digital or hard copies of all or part of this work for personal or

classroom use is granted without fee provided that copies are not made or distributed

for proﬁt or commercial advantage and that copies bear this notice and the full cita-

tion on the ﬁrst page. Copyrights for components of this work owned by others than

ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or re-

publish, to post on servers or to redistribute to lists, requires prior speciﬁc permission

and/or a fee. Request permissions from permissions@acm.org.

ICCDA ’17 May 19-23, 2017, Lakeland, FL, USA

 2017 Association for Computing Machinery.

ACM ISBN 978-1-4503-5241-3/17/05. .. $15.00

http://dx.doi.org/10.1145/3093241.3093278

large-scale production applications use RocksDB to manage stor-

age, leveraging its high performance to mitigate the ever-growing

pressure on the storage-system [4].

Unfortunately, RocksDB ﬂexibility and superior performance come

at a cost: tuning RocksDB is a complex task that involves more than

a hundred parameters with varying levels of inter-dependencies.

Furthermore, “while recent changes have made RocksDB better, it

is much harder to conﬁgure than LevelDB”; too often poor results

“are caused by misconﬁguration” [5].

The main questions raised when operating with RocksDB are: (1)

which conﬁguration parameters should be used for which hardware

and under what workload? (2) what are the optimal values for these

parameters? (3) are parameters interdependent (i.e., tuning param-

eter a works if and only if parameters b, c and d have certain val-

ues)? (4) will the positive optimization from two different tunings

cumulate or negate when brought together? Last but not least, (5)

what, if any, are the side effects of these optimizations?

This paper seeks to answer these questions by sharing our ex-

perience optimizing RocksDB in the context of Redis-on-Flash

(RoF) [6, 7] – a commercial extension to the popular Redis in-

memory key value store [8]. RoF uses SSDs as a RAM extension

to provide competitive performance to the in-memory Redis vari-

ant while dramatically increasing the effective dataset capacity that

can be stored on a single server. In RoF, hot values are saved in

RAM, while cold-values are saved in SSDs and are managed by

RocksDB (See Section 2.2). Because RocksDB handles all of RoF

accesses to storage, its performance plays a major role in the over-

all system performance, especially for use cases with low access

locality. Since RoF aims to provide competitive performance to

the pure-RAM Redis variant, tuning RocksDB proved to be a key

challenge.

During the process of tuning RocksDB for the RoF case, we ana-

lyzed a large set of parameters and experimented with their impact

on the performance for several different workloads – database repli-

cation, a write-only workload, and a 50-50 read:write workload.

To verify the robustness of our settings across different hardware

setups, we run all experiments in both Amazon Elastic Compute

Cloud (EC2) and Google Compute Engine (GCE). Overall, our tun-

ing reduced the time needed to replicated a node by more than 11x.

The bulk of this paper describes the methodology, tuning process,

and speciﬁc parameters settings that lead us to this result.

In Section 3, we describe our methodology and explain the experi-

ments process. Then, in Section 4 we detail the parameters tuning

that had the largest positive effect on performance. We also specif-

ically list parameters for which we expected performance improve-

ment but instead either reduced performance or had other negative

下载后可阅读完整内容，剩余6页未读，立即下载

边城水手

粉丝: 113
资源: 35

优化RocksDB以提升Redis在闪存上的性能

广义q-ROF TODIM决策方法研究与应用分析

Kramers-Kronig接收器在SSB-OFDM-RoF链路的应用与优化

OFCG中多重光载波抑制的WDM-RoF光上转换技术

RoF.zip_Links_OFDM-ROF_Radio over fiber_evm ofdm_ofdm

ROF-Model-FFT-Transform.rar_rof

基于偏振复用和反射式半导体光放大器的WDM-RoF-PON系统设计.pdf

ROF-system.rar_ROF系统_rof_光载无线_光载无线通信_光通信

网络技术-网络基础-ROF无线接入技术研究.pdf

论文研究-ROF系统中毫米波生成技术的研究 .pdf

论文研究-ROF系统中毫米波光学生成方法的研究 .pdf

最新资源