Article Preview
Top1. Introduction
Cloud Service Providers (CSPs) are increasing their services and infrastructures drastically because of huge scope and many benefits of cloud computing technology (Birje, Challagidad, Goudar, & Tapale, 2017; Birje & Challagidad, 2015). Hosting of number of services (such as conventional applications and strong consistency services) in cloud infrastructures is accelerating day-by-day (Gummadi, 2004), (Haq, 2017).
In conventional methods, the application states are replicated over many servers spread in different geographical locations across the world to improve the availability of the application state. Replicating state across data centers, however, makes it harder to maintain consistency across updates (Anderson, Meling, Rasmussen, Vahdat, & Marzullo, 2017; DeCandia, 2007). To have correct performance of the system for many cloud-oriented services strong consistency maintenance is essential. Strongly consistent cloud services must provide good performance and high availability. Content replication over multiple servers/data centers is one of the conventional methods used to achieve both the goals. Many services of cloud gain advantage from the simplicity and semantics of strong consistency. However, conventional strong consistency replication methods have the limitation of availability of replicated services when recovering huge amount of data across wide area links (Anderson, Meling, Rasmussen, Vahdat, & Marzullo, 2017). The necessity of strong consistency is also identified by prominent CSPs (Calder et. al., 2011; Corbett, 2012).
It is difficult to provide strong consistency for data residing in multiple sites because of dissimilar updates probably from various users, may be aimed at servers in various data centers (Anderson, Meling, Rasmussen, Vahdat, & Marzullo, 2017). Though updates are moved over the similar server, the arrays in which numerous updates are dedicated to application state should be diagonally similar across all replicas of the state. A Replicated State Machine (RSM) model (Schneider, 1990) typically achieves strong consistency, using consensus algorithm such as Paxos (Lamport, 1998; Paxos, 2001; Taft, 2009; Chandra, 2007) to direct state machine operations. In this context, an elementary requirement for strong consistency in RSMs is 2f + 1 replicas are required to tolerate f failures. Furthermore, the real-world consists of billions of items (Taft, 2009; Chandra, 2007) stored over several data centers; every datacenter consists of millions of machines need to have local recovery and remote recovery mechanisms. For these huge networks of machines, recovery from common machine failures continuously is the requirement without human intervention.