Analysis for REPERA: A Hybrid Data Protection Mechanism in Distributed Environment

Analysis for REPERA: A Hybrid Data Protection Mechanism in Distributed Environment

Longbin Lai (Shanghai Jiao Tong University, China), Linfeng Shen (IBM China Development Lab, China), Yanfei Zheng (Shanghai Jiao Tong University, China), Kefei Chen (Shanghai Jiao Tong University, China) and Jing Zhang (Shanghai Jiao Tong University, China)
Copyright: © 2012 |Pages: 12
DOI: 10.4018/ijcac.2012010105
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

Distributed systems, especially those providing cloud services, endeavor to construct sufficiently reliable storage in order to attract more customers. Generally, pure replication and erasure code are widely adopted in distributed systems to guarantee reliable data storage, yet both of them contain some deficiencies. Pure replication consumes too much extra storage and bandwidth, while erasure code seems not so high-efficiency and only suitable for read-only context. The authors proposed REPERA as a hybrid mechanism combining pure replication and erasure code to leverage their advantages and mitigate their shortages. This paper qualitatively compares fault-resilient distributed architectures built with pure replication, erasure code and REPERA. The authors show that systems employing REPERA share with erasure-resilient systems a higher availability and more durable storage with similar space and bandwidth consumption when compared with replicated systems. The authors show that systems employing REPERA, on one hand, obtain higher availability while comparing to erasure-resilient systems, on the other hand, benefit from more durable storage while comparing to replicated systems. Furthermore, since REPERA was developed under the open platform, REPERA, the authors prepare an experiment to evaluate the performance of REPERA by comparing with the original system.
Article Preview

1. Introduction

Cloud has now attracted both commercial and academic focus. Cloud sophisticatedly combines the economic, commercial, cultural and technological conditions, causing a disruptive shift in IT industry towards a service-based economy. Usually, large distributed systems, especially those providing services as cloud, are built upon cheap hardware to reduce cost and are exposed to the Internet to accumulate customers accompanied with their massive information. However, cheap components follow by potential hardware failures, while public exposure leads to the facilities of hack attacks and viruses. Furthermore more, disk drives are statistically the most commonly replaced hardware components in large storage systems, accounting for almost 30% to 50% of all hardware replacements (Schroeder & Gibson, 2007). Thereby, mechanism providing reliable storage turns out as one of the most essential components in design of distributed systems.

Generally, pure replication and erasure code are introduced to guarantee highly-available storage, and they obtain widespread adoption in distributed systems. Google File System (Ghemawat, Gobioff, & Shun-Tak, 2003), Pastry (Rowstron & Druschel, 2007), and Tapestry (Zhao, Kubiatowicz, & Joseph, 2001) are typical replicated systems, in which data is mirroring not only to tolerate failures, but also to support concurrent data access in consideration of latency reduction. However, pure replication consumes too much extra storage and bandwidth, hence boosting the cost for deploying additional devices and managements. Erasure code has then been introduced into storage systems to increase storage durability (also known as the expected mean time to failure, or MTTF, of losing any data unit) and to reduce the investment of preparing massive storage as well as the extra transferring and maintaining overhead for the replicas. The key idea behind erasure code is that m blocks of source data are encoded to produce n blocks of encoded data, in such a way that any subset of m encoded blocks suffices to reconstruct the source data. Such a code is called an (n, m) erasure code and allows up to losses in a group of n encoded blocks (Rizzo, 1997). Considering a data sized , 3-way replication consumes storage space as well as transferring bandwidth, while (7, 4) erasure code only takes up nearly 7/4 , to suffer up to 3-disk failure. Chen, Edler, Goldberg, Gottlieb, Sobti, and Yianilos (1999) initially implemented erasure code in the distributed context to build a high-efficient archival intermemory. Aguilera and Janakiraman (2005) presented methods of efficiently using erasure code in distributed systems. Weatherspoon and Kubiatowicz (2002) pointed out that systems employing erasure code have MTTF many orders of magnitude higher than replicated systems with similar storage and bandwidth requirements, and reversely, erasure-resilient systems take up much less bandwidth and storage to provide similar system durability as replicated systems. However, besides an extra computation for coding procedure, erasure coding mechanism disappoints system designers for a longer access delay as multiple servers containing the erasure stripes have to be contacted to read a single block and it encounters bottlenecks when either modifications or failures happen frequently, which may introduce excessive coding processes as to possibly drain the system.

Complete Article List

Search this Journal:
Reset
Open Access Articles: Forthcoming
Volume 7: 4 Issues (2017)
Volume 6: 4 Issues (2016)
Volume 5: 4 Issues (2015)
Volume 4: 4 Issues (2014)
Volume 3: 4 Issues (2013)
Volume 2: 4 Issues (2012)
Volume 1: 4 Issues (2011)
View Complete Journal Contents Listing