FEC4Cloud: a research project promoting erasure coding for Cloud storage architectures - Inria - Institut national de recherche en sciences et technologies du numérique Accéder directement au contenu
Communication Dans Un Congrès Année : 2015

FEC4Cloud: a research project promoting erasure coding for Cloud storage architectures

Résumé

The aim of FEC4Cloud project is to promote erasure codes in fault-tolerant Cloud storage (and computing) architectures. For this purpose, replication is commonly used to maintain the availability rate within the system. But, replication i.e triplication is storage capacity consuming. A network area of storage need usually two times more space of storage with replication than with erasure code approaches for the same level of protection (no more than four erasures basically). Intuitively, the complexity of erasure code could be opposed to real time read/write of I/O intensive application (with hot data). That leads to erasure code for archiving application traditionally (with some cold data). In this talk, we want to demonstrate (practically) that some erasure code can be used both for cold and hot data accesses. In FEC4Cloud, we investigate, in parallel, algebraic approach like the Reed-Solomon codes and discrete geometric approach supported notably by the Mojette transform (a discrete and exact Radon transform that will be presented). At this time, the implementations based on Mojette transform provide the best throughput (in MB/s) in encoding and decoding but it is possible (in very particular configurations) to accelerate highly the classical algebraic approach. In this talk, we will provide a comparison with coding schemes of [1] and [4] with a last update of results presented in [2] that brings a huge advantage to discrete geometric approaches (up to 6 time faster of reference Reed-Solomon coding schemes recently optimized in ISA library of Intel). The talk will also include results on several experimentations of a Mojette erasure-coded distributed file system (called RozoFS). In particular, we will focus on a real testbed that provides video editing on line service based on RozoFS to show that it is possible to use erasure code not only for archiving but also for I/O intensive applications [3]. Last results on privacy in cloud storage architecture will conclude the talk.
Fichier non déposé

Dates et versions

hal-01156462 , version 1 (05-06-2015)

Identifiants

  • HAL Id : hal-01156462 , version 1

Citer

Benoît Parrein, Jérôme Lacan, Nicolas Normand, Dimitri Pertin, Jonathan Detchart, et al.. FEC4Cloud: a research project promoting erasure coding for Cloud storage architectures. Rendez-vous de la Recherche et de l'Enseignement de la Sécurité des Systèmes d'Information (RESSI), May 2015, Troyes, France. ⟨hal-01156462⟩
253 Consultations
1 Téléchargements

Partager

Gmail Facebook X LinkedIn More