Browse Prior Art Database

RELIABLE, SCALABLE, AND HIGH-­PERFORMANCE DISTRIBUTED STORAGE: Distributed Object Storage

IP.com Disclosure Number: IPCOM000234957D
Original Publication Date: 2014-Feb-19
Included in the Prior Art Database: 2014-Feb-19

Publishing Venue

Linux Defenders

Related People

Sage Weil: AUTHOR

Abstract

Distributed object storage architecture leverages device intelligence to provide a reliable and scalable storage abstraction with minimal oversight. Described is an efficient, scalable, and low ­overhead cluster management protocol that facilitates consistent and coherent data access through the propagation of small cluster maps that specify device membership and data distribution. This allows a dynamic cluster of semi­-autonomous OSDs to self­ manage consistent data replication, failure detection, and failure recovery while providing the illusion of a single logical object store with direct, high­ performance client access to data.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 9% of the total text.

Page 01 of 10

RELIABLE, SCALABLE, AND HIGH­PERFORMANCE DISTRIBUTED STORAGE: Distributed Object Storage

Authored by: Sage Weil

Abstract

Distributed object storage architecture leverages device intelligence to provide a reliable and scalable storage abstraction               
with minimal oversight. Described is an efficient, scalable, and low­overhead cluster management protocol that facilitates

                  consistent and coherent data access through the propagation of small cluster maps that specify device membership and

                      
data distribution. This allows a dynamic cluster of semi­autonomous OSDs to self­manage consistent data replication,

        

                                                
failure detection, and failure recovery while providing the illusion of a single logical object store with direct,                  high­performance client access to data.

Keywords: RADOS, cluster map, data distribution, object storage devices

Introduction

The work described is a scalable and reliable object storage service termed RADOS (Reliable, Autonomic Distributed

                  Object Store) without compromising performance. RADOS fa​

cilitates an evolving, balanced distribution of data and

   

                                 workload across a dynamic and hetero​

geneous storage cluster while providing applications with the illusion of a single

                      logical object store with well­defined safety semantics and strong consistency guarantees. Metadata bottle​

necks associated

                 
with data layout and storage allocation are eliminated through the use of a compact cluster map that describes cluster state                     and data layout in terms of placement groups.

By separating serialization from safety, the architecture provides strong consis​

tency semantics to applications by minimally

                 involving clients in failure recovery. RADOS utilizes a globally replicated cluster map that provides all parties with

                       com​

plete knowledge of the data distribution, typically specified using a function like CRUSH. This avoids the need for

                       object lookup present in conventional architectures, which RADOS lever​

ages to distribute replication, consistency

                 management, and failure recovery among a dynamic cluster of OSDs while still preserving consistent read and update

                         semantics. A scalable failure detection and cluster map distribution strategy enables the creation of extremely large storage

                       clusters, with minimal oversight by the tightly­coupled and highly reliable monitor cluster that manages the master copy of

      

                                            the map. Because clusters at the petabyte scale are necessarily...