Designing a Storage Software Stack for Accelerators

Shinichi Awamoto; Erich Focht; Michio Honda

Designing a Storage Software Stack for Accelerators

Shinichi Awamoto, Erich Focht, Michio Honda

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

Abstract

Although modern accelerator devices, such as vector engines and SmartNICs, are equipped with general purpose CPUs, access to the storage needs the mediation of the host kernel and CPUs, resulting in latency and throughput penalties. In this paper, we explore the case for direct storage access inside the accelerator applications, and discuss the problem, design options and benefits of this architecture. We demonstrate that our architecture can improve throughputs of LevelDB by 12–89%, and reduce the execution time by 33–46 % in a bioinformatics application in comparison to the baseline where the host system mediates the storage accesses.

Original language	English
Title of host publication	Proceedings of the 12th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage 20)
Publisher	USENIX Association
Number of pages	7
Publication status	Published - 13 Jul 2020
Event	12th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage 20) - virtual Duration: 13 Jul 2020 → 14 Jul 2020 https://www.usenix.org/conference/hotstorage20

Workshop

Workshop	12th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage 20)
Period	13/07/20 → 14/07/20
Internet address	https://www.usenix.org/conference/hotstorage20

Access to Document

https://www.usenix.org/conference/hotstorage20/presentation/awamotoLicence: Unspecified

Fingerprint Dive into the research topics of 'Designing a Storage Software Stack for Accelerators'. Together they form a unique fingerprint.

Cite this

@inproceedings{57b8226da55f4fb695e4de0107227081,

title = "Designing a Storage Software Stack for Accelerators",

abstract = "Although modern accelerator devices, such as vector engines and SmartNICs, are equipped with general purpose CPUs, access to the storage needs the mediation of the host kernel and CPUs, resulting in latency and throughput penalties. In this paper, we explore the case for direct storage access inside the accelerator applications, and discuss the problem, design options and benefits of this architecture. We demonstrate that our architecture can improve throughputs of LevelDB by 12–89%, and reduce the execution time by 33–46 % in a bioinformatics application in comparison to the baseline where the host system mediates the storage accesses.",

author = "Shinichi Awamoto and Erich Focht and Michio Honda",

year = "2020",

month = jul,

day = "13",

language = "English",

booktitle = "Proceedings of the 12th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage 20)",

publisher = "USENIX Association",

note = "12th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage 20) ; Conference date: 13-07-2020 Through 14-07-2020",

url = "https://www.usenix.org/conference/hotstorage20",

}

Awamoto, S, Focht, E & Honda, M 2020, Designing a Storage Software Stack for Accelerators. in Proceedings of the 12th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage 20). USENIX Association, 12th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage 20), 13/07/20. <https://www.usenix.org/conference/hotstorage20/presentation/awamoto>

TY - GEN

T1 - Designing a Storage Software Stack for Accelerators

AU - Awamoto, Shinichi

AU - Focht, Erich

AU - Honda, Michio

PY - 2020/7/13

Y1 - 2020/7/13

N2 - Although modern accelerator devices, such as vector engines and SmartNICs, are equipped with general purpose CPUs, access to the storage needs the mediation of the host kernel and CPUs, resulting in latency and throughput penalties. In this paper, we explore the case for direct storage access inside the accelerator applications, and discuss the problem, design options and benefits of this architecture. We demonstrate that our architecture can improve throughputs of LevelDB by 12–89%, and reduce the execution time by 33–46 % in a bioinformatics application in comparison to the baseline where the host system mediates the storage accesses.

AB - Although modern accelerator devices, such as vector engines and SmartNICs, are equipped with general purpose CPUs, access to the storage needs the mediation of the host kernel and CPUs, resulting in latency and throughput penalties. In this paper, we explore the case for direct storage access inside the accelerator applications, and discuss the problem, design options and benefits of this architecture. We demonstrate that our architecture can improve throughputs of LevelDB by 12–89%, and reduce the execution time by 33–46 % in a bioinformatics application in comparison to the baseline where the host system mediates the storage accesses.

M3 - Conference contribution

BT - Proceedings of the 12th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage 20)

PB - USENIX Association

T2 - 12th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage 20)

Y2 - 13 July 2020 through 14 July 2020

ER -