Erasure coded storage system based database as a service with multi level consistency support

Abstract

Today s world is highly data-driven, with data of various forms being generated across the world by varying sources. Social media, mobile phones, sensors, online shopping sites, financial transactions, analysis of human genomes, and many others play significant roles as contributors of data. The generated data has critical information or patterns hidden in it that helps business houses and researchers to analyze and make decisions about their future action. Efficient storage of the data generated is an essential step in the analysis of the data, and cloud-based storage service providers play a significant role in this step. Availability, resilience, and mobility are the major characteristics that make cloud storage attractive. Data replication is the primary method adopted by cloud storage service providers to ensure availability. Replication of data makes multiple copies of the same data to ensure availability even in the context of failure of one copy of the data. The standard is to maintain three copies of data so that the failure of up to two copies does not affect availability. In this era of Big Data, the volume of data generated is so high that the replication of data for fault tolerance leads to huge storage overhead. An alternative method to solve this storage overhead issue is Erasure-Coded storage. Erasure-Coded storage provides the same fault tolerance as data replication based storage, but at a reduced storage overhead. Many leading cloud storage service providers have already started using this method to store their warm data (less frequently accessed data). Cloud storage service providers use Erasure-Coded storage systems for storing large volumes of data. In this context, Erasure-Coded storage systems, in which storage node failure is a common occurrence, require mechanisms to identify the efficient set of nodes to replace the failed nodes. newline