東華大學圖書館 |

Language: English

Help

回圖書館首頁

手機版館藏查詢

Back

Switch To: Labeled | MARC Mode | ISBD

Enhancing I/O Observability and Stor...

Han, Runzhou,

Linked to FindBook

Google Book

Amazon

博客來

Enhancing I/O Observability and Storage Reliability in High-Performance Computing Systems /

Record Type:	Electronic resources : Monograph/item
Title/Author:	Enhancing I/O Observability and Storage Reliability in High-Performance Computing Systems // Runzhou Han.
Author:	Han, Runzhou,
Description:	1 electronic resource (173 pages)
Notes:	Source: Dissertations Abstracts International, Volume: 85-12, Section: A.
Contained By:	Dissertations Abstracts International85-12A.
Subject:	Computer engineering. -
Online resource:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31235745
ISBN:	9798382801506

Enhancing I/O Observability and Storage Reliability in High-Performance Computing Systems /
Han, Runzhou,

Enhancing I/O Observability and Storage Reliability in High-Performance Computing Systems /Runzhou Han. - 1 electronic resource (173 pages)

Source: Dissertations Abstracts International, Volume: 85-12, Section: A.

High-performance computing (HPC) plays an essential role in scientific research, engineering, and business. Due to the complexity of HPC, analysis of HPC workload and HPC system is a both critical and challenging to the community.We first introduce a data-centric provenance solution to enhance observability of HPC workloads. Data provenance captures the life cycle of data, including its origins and usage patterns in scientific workflows on HPC systems. Existing provenance solutions are inadequate due to incompatible models and system implementations. We collaborate with domain scientists to survey HPC workflows and identify specific provenance requirements. Based on the survey, we propose PROV-IO+, a provenance framework that incorporates an I/O-centric model to precisely describe scientific data, associated I/O operations, and environments. Our experiments with real-world workflows demonstrate that PROV-IO+ effectively meets the provenance needs of domain scientists with reasonable performance.Then, we introduce our study of the failure recovery and logging mechanisms of parallel file systems (PFSes) which are a major subset of high-performance distributed storage systems (DSSes). We propose a tool called PFault to simulate failures and analyze the behavior of PFSes. We focus on two widely used PFSes (Lustre and BeeGFS) and find areas where they can improve in handling failures. We also investigate the causes of abnormalities, and our analysis has led to a patch for Lustre. Additionally, we characterize the extensive logs generated in the experiments in details and identify the unique patterns and limitations of PFSes in terms of failure logging. Moreover, we further study the capabilities of PFS checkers with more fine-gained fault injection and identify instances of sub-optimal behavior leading to issues like kernel panics and incorrect repairs.Finally, we present our effort in studying erasure coding (EC) performance in high-performance storage systems. Erasure coding plays a crucial role in the fault tolerance of modern high-performance DSSes. Inspired by recent research on storage configuration, we study the configuration sensitivity of EC in real DSS in this work. Our results show that configurations may affect the EC recovery time significantly, and EC may introduce extra write amplification (WA) than the theoretical expectation. Our work suggests the importance of considering the context of real DSS for EC research and implementation.

English

ISBN: 9798382801506Subjects--Topical Terms:

621879
Computer engineering.
Subjects--Index Terms:

High-performance computing

Enhancing I/O Observability and Storage Reliability in High-Performance Computing Systems /
LDR:03991nmm a22004453i 4500 001 2432377
005 20250707081451.5
006 m o d
007 cr|nu||||||||
008 261011s2024 miu||||||m |||||||eng d
020 $a 9798382801506
035 $a (MiAaPQD)AAI31235745
035 $a AAI31235745
040 $a MiAaPQD $b eng $c MiAaPQD $e rda
100 1 $a Han, Runzhou, $e author. $3 3821575
245 1 0 $a Enhancing I/O Observability and Storage Reliability in High-Performance Computing Systems / $c Runzhou Han.
264 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2024
300 $a 1 electronic resource (173 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 85-12, Section: A.
500 $a Advisors: Zheng, Mai Committee members: Guan, Yong; Gulmezoglu, Berk; Jacobson, Doug; Ramamoorthy, Aditya.
502 $b Ph.D. $c Iowa State University $d 2024.
520 $a High-performance computing (HPC) plays an essential role in scientific research, engineering, and business. Due to the complexity of HPC, analysis of HPC workload and HPC system is a both critical and challenging to the community.We first introduce a data-centric provenance solution to enhance observability of HPC workloads. Data provenance captures the life cycle of data, including its origins and usage patterns in scientific workflows on HPC systems. Existing provenance solutions are inadequate due to incompatible models and system implementations. We collaborate with domain scientists to survey HPC workflows and identify specific provenance requirements. Based on the survey, we propose PROV-IO+, a provenance framework that incorporates an I/O-centric model to precisely describe scientific data, associated I/O operations, and environments. Our experiments with real-world workflows demonstrate that PROV-IO+ effectively meets the provenance needs of domain scientists with reasonable performance.Then, we introduce our study of the failure recovery and logging mechanisms of parallel file systems (PFSes) which are a major subset of high-performance distributed storage systems (DSSes). We propose a tool called PFault to simulate failures and analyze the behavior of PFSes. We focus on two widely used PFSes (Lustre and BeeGFS) and find areas where they can improve in handling failures. We also investigate the causes of abnormalities, and our analysis has led to a patch for Lustre. Additionally, we characterize the extensive logs generated in the experiments in details and identify the unique patterns and limitations of PFSes in terms of failure logging. Moreover, we further study the capabilities of PFS checkers with more fine-gained fault injection and identify instances of sub-optimal behavior leading to issues like kernel panics and incorrect repairs.Finally, we present our effort in studying erasure coding (EC) performance in high-performance storage systems. Erasure coding plays a crucial role in the fault tolerance of modern high-performance DSSes. Inspired by recent research on storage configuration, we study the configuration sensitivity of EC in real DSS in this work. Our results show that configurations may affect the EC recovery time significantly, and EC may introduce extra write amplification (WA) than the theoretical expectation. Our work suggests the importance of considering the context of real DSS for EC research and implementation.
546 $a English
590 $a School code: 0097
650 4 $a Computer engineering. $3 621879
650 4 $a Computer science. $3 523869
650 4 $a Systems science. $3 3168411
650 4 $a Information science. $3 554358
653 $a High-performance computing
653 $a Scientific data management
653 $a Storage systems
653 $a Parallel file systems
653 $a Erasure coding
690 $a 0464
690 $a 0984
690 $a 0723
690 $a 0790
710 2 $a Iowa State University. $b Electrical and Computer Engineering. $e degree granting institution. $3 3821576
720 1 $a Zheng, Mai $e degree supervisor.
773 0 $t Dissertations Abstracts International $g 85-12A.
790 $a 0097
791 $a Ph.D.
792 $a 2024
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31235745