SplinterDB: Closing the bandwidth gap for NVMe key-value stores

Alex Conway; Abhishek Gupta; Vijay Chidambaran; Martin Farach-Colton; Rick Spillane; Amy Tai; Rob Johnson

SplinterDB: Closing the bandwidth gap for NVMe key-value stores

Alex Conway, Abhishek Gupta, Vijay Chidambaran, Martin Farach-Colton, Rick Spillane, Amy Tai, Rob Johnson

School of Arts and Sciences, Computer Science

Rutgers, The State University

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

4 Scopus citations

Abstract

Modern NVMe solid state drives offer significantly higher bandwidth and lower latency than prior storage devices. Current key-value stores struggle to fully utilize the bandwidth of such devices. This paper presents SplinterDB, a new key-value store explicitly designed for NVMe solid-state-drives. SplinterDB is designed around a novel data structure (the STB^ε-tree) that exposes I/O and CPU concurrency and reduces write amplification without sacrificing query performance. STB^ε-tree combines ideas from log-structured merge trees and B^ε-trees to reduce write amplification and CPU costs of compaction. The SplinterDB memtable and cache are designed to be highly concurrent and to reduce cache misses. We evaluate SplinterDB on a number of micro- and macro-benchmarks, and show that SplinterDB outperforms RocksDB, a state-of-the-art key-value store, by a factor of 6-10× on insertions and 2-2.6× on point queries, while matching RocksDB on small range queries. Furthermore, SplinterDB reduces write amplification by 2× compared to RocksDB.

Original language	American English
Title of host publication	Proceedings of the 2020 USENIX Annual Technical Conference, ATC 2020
Publisher	USENIX Association
Pages	49-63
Number of pages	15
ISBN (Electronic)	9781939133144
State	Published - 2020
Event	2020 USENIX Annual Technical Conference, ATC 2020 - Virtual, Online Duration: Jul 15 2020 → Jul 17 2020

Publication series

Name	Proceedings of the 2020 USENIX Annual Technical Conference, ATC 2020

Conference

Conference	2020 USENIX Annual Technical Conference, ATC 2020
City	Virtual, Online
Period	7/15/20 → 7/17/20

ASJC Scopus subject areas

General Computer Science

Cite this

@inproceedings{0403c77c95ff4e76930fd6173d947b3b,

title = "SplinterDB: Closing the bandwidth gap for NVMe key-value stores",

abstract = "Modern NVMe solid state drives offer significantly higher bandwidth and lower latency than prior storage devices. Current key-value stores struggle to fully utilize the bandwidth of such devices. This paper presents SplinterDB, a new key-value store explicitly designed for NVMe solid-state-drives. SplinterDB is designed around a novel data structure (the STBε-tree) that exposes I/O and CPU concurrency and reduces write amplification without sacrificing query performance. STBε-tree combines ideas from log-structured merge trees and Bε-trees to reduce write amplification and CPU costs of compaction. The SplinterDB memtable and cache are designed to be highly concurrent and to reduce cache misses. We evaluate SplinterDB on a number of micro- and macro-benchmarks, and show that SplinterDB outperforms RocksDB, a state-of-the-art key-value store, by a factor of 6-10× on insertions and 2-2.6× on point queries, while matching RocksDB on small range queries. Furthermore, SplinterDB reduces write amplification by 2× compared to RocksDB.",

author = "Alex Conway and Abhishek Gupta and Vijay Chidambaran and Martin Farach-Colton and Rick Spillane and Amy Tai and Rob Johnson",

note = "Publisher Copyright: Copyright {\textcopyright} Proc. of the 2020 USENIX Annual Technical Conference, ATC 2020. All rights reserved.; 2020 USENIX Annual Technical Conference, ATC 2020 ; Conference date: 15-07-2020 Through 17-07-2020",

year = "2020",

language = "American English",

series = "Proceedings of the 2020 USENIX Annual Technical Conference, ATC 2020",

publisher = "USENIX Association",

pages = "49--63",

booktitle = "Proceedings of the 2020 USENIX Annual Technical Conference, ATC 2020",

}

Conway, A, Gupta, A, Chidambaran, V, Farach-Colton, M, Spillane, R, Tai, A & Johnson, R 2020, SplinterDB: Closing the bandwidth gap for NVMe key-value stores. in Proceedings of the 2020 USENIX Annual Technical Conference, ATC 2020. Proceedings of the 2020 USENIX Annual Technical Conference, ATC 2020, USENIX Association, pp. 49-63, 2020 USENIX Annual Technical Conference, ATC 2020, Virtual, Online, 7/15/20.

SplinterDB: Closing the bandwidth gap for NVMe key-value stores. / Conway, Alex; Gupta, Abhishek; Chidambaran, Vijay et al.
Proceedings of the 2020 USENIX Annual Technical Conference, ATC 2020. USENIX Association, 2020. p. 49-63 (Proceedings of the 2020 USENIX Annual Technical Conference, ATC 2020).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - SplinterDB

T2 - 2020 USENIX Annual Technical Conference, ATC 2020

AU - Conway, Alex

AU - Gupta, Abhishek

AU - Chidambaran, Vijay

AU - Farach-Colton, Martin

AU - Spillane, Rick

AU - Tai, Amy

AU - Johnson, Rob

PY - 2020

Y1 - 2020

N2 - Modern NVMe solid state drives offer significantly higher bandwidth and lower latency than prior storage devices. Current key-value stores struggle to fully utilize the bandwidth of such devices. This paper presents SplinterDB, a new key-value store explicitly designed for NVMe solid-state-drives. SplinterDB is designed around a novel data structure (the STBε-tree) that exposes I/O and CPU concurrency and reduces write amplification without sacrificing query performance. STBε-tree combines ideas from log-structured merge trees and Bε-trees to reduce write amplification and CPU costs of compaction. The SplinterDB memtable and cache are designed to be highly concurrent and to reduce cache misses. We evaluate SplinterDB on a number of micro- and macro-benchmarks, and show that SplinterDB outperforms RocksDB, a state-of-the-art key-value store, by a factor of 6-10× on insertions and 2-2.6× on point queries, while matching RocksDB on small range queries. Furthermore, SplinterDB reduces write amplification by 2× compared to RocksDB.

AB - Modern NVMe solid state drives offer significantly higher bandwidth and lower latency than prior storage devices. Current key-value stores struggle to fully utilize the bandwidth of such devices. This paper presents SplinterDB, a new key-value store explicitly designed for NVMe solid-state-drives. SplinterDB is designed around a novel data structure (the STBε-tree) that exposes I/O and CPU concurrency and reduces write amplification without sacrificing query performance. STBε-tree combines ideas from log-structured merge trees and Bε-trees to reduce write amplification and CPU costs of compaction. The SplinterDB memtable and cache are designed to be highly concurrent and to reduce cache misses. We evaluate SplinterDB on a number of micro- and macro-benchmarks, and show that SplinterDB outperforms RocksDB, a state-of-the-art key-value store, by a factor of 6-10× on insertions and 2-2.6× on point queries, while matching RocksDB on small range queries. Furthermore, SplinterDB reduces write amplification by 2× compared to RocksDB.

UR - http://www.scopus.com/inward/record.url?scp=85091906557&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85091906557&partnerID=8YFLogxK

M3 - Conference contribution

T3 - Proceedings of the 2020 USENIX Annual Technical Conference, ATC 2020

SP - 49

EP - 63

BT - Proceedings of the 2020 USENIX Annual Technical Conference, ATC 2020

PB - USENIX Association

Y2 - 15 July 2020 through 17 July 2020

ER -

SplinterDB: Closing the bandwidth gap for NVMe key-value stores

Abstract

Publication series

Conference

ASJC Scopus subject areas

Other files and links

Fingerprint

Cite this