Descripción
|
|
---|---|
Small ?les are known to pose major performance challenges for ?le systems. Yet, such workloads are increasingly common in a number of Big Data Analytics work?ows or large- scale HPC simulations. These challenges are mainly caused by the common architecture of most state-of-the-art ?le systems needing one or multiple metadata requests before being able to read from a ?le. Small input ?le size causes the overhead of this metadata management to gain relative importance as the size of each ?le decreases. In this paper we propose a set of techniques leveraging consistent hashing and dynamic metadata replication to signi?cantly reduce this metadata overhead. We implement such techniques inside a new ?le system named T¿ yrFS, built as a thin layer above the T¿ yr object store. We prove that T¿ yrFS increases small ?le access performance up to one order of magnitude compared to other state-of-the-art ?le systems, while only causing a minimal impact on ?le write throughput. | |
Internacional
|
Si |
Nombre congreso
|
18th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing, CCGrid 2018 |
Tipo de participación
|
960 |
Lugar del congreso
|
Washington, Estados Unidos de América |
Revisores
|
Si |
ISBN o ISSN
|
0-7695-6410-0 |
DOI
|
10.1109/CCGRID.2018.00072 |
Fecha inicio congreso
|
01/05/2018 |
Fecha fin congreso
|
04/05/2018 |
Desde la página
|
452 |
Hasta la página
|
461 |
Título de las actas
|
CCGrid 2018, Proceedings of the 18th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing, CCGrid 2018. |