Decentralized storage systems are growing far more attractive over the years, as it is being employed by the popular networking corporations like Facebook [Cassandra], Amazon [Dynamo], LinkedIn [Voldemort] and etc. Behind the growing popularity is the rising number of researches and developments in the field of decentralized storage systems. The basic defining feature of decentralized storage systems is a system, in which all nodes have same capabilities, responsibilities, and have symmetric communications [past]. It is in contrast with centralized storage system where the data is stored in one or few dedicated servers.

In this paper we treat decentralized storage systems as same as peer-to-peer storage systems, due to their practical needs as storage for volunteer computing systems. Hence, we assume the same characteristics of P2P storage systems in the decentralized storage system.

Adroutsellis-Theotokis et al. defined a “litmus test” for P2P systems by asking the following questions:

The goals of decentralized storage system are an ad hoc connectivity, a decentralization of the system, reduced cost of ownership, and anonymity.

In this paper we surveyed number of different decentralized storage systems with their key features and pros-and-cons, as well as evaluated them according to their suitableness in storage system for volunteer computing. The characteristics that are important for storage system in volunteer computing are fault tolerance, reliability, availability and scalability.

The design issues that associate with these characteristics are implementation of read/write access, replication handling among nodes in the system for fast resource location, symmetry design of the storage for load balancing, fault and security handling for reliable storage, and the scalability feature for handling thousands of nodes. Multi-writer designs face numerable issues not found in read-only systems, such as maintaining consistencies among replicas, handling concurrent updates, and preventing malicious attacks. Similarly, one can find multiple approaches in replications within the storage systems, for high reliability, churn protection, and fast resource location. Same goes to fault tolerance, there are multiple techniques to achieve certain tolerance level against faulty nodes. For example Dynamo’s sloppy quorum and hinted handoff as well as Total Recall’s redundancy management. In addition, security is equally important in peer-to-peer storage systems for volunteer computing, considering the importance of data integrity and computing resources of the volunteers in volunteer computing systems.

decentralized_storage_systems/dss_background.txt · Last modified: 2012/04/23 00:50 by julia
 
Except where otherwise noted, content on this wiki is licensed under the following license: CC Attribution-Share Alike 3.0 Unported
Recent changes RSS feed Donate Powered by PHP Valid XHTML 1.0 Valid CSS Driven by DokuWiki