From Byzantine agreement to practical survivability: a position paper

Only a decade ago, issues of replication, high availability and load balancing were the focus of small, closely coupled cluster projects. Consequently, techniques for cluster management and small replication systems are abundant. However, the advent of the Internet led to wide spread and highly decentralized access of services and content that bring issues of scale and ubiquitous deployment. In particular, the need to maintain copies of replicated data consistent grows beyond the limits of any local cluster. Consequently, researchers have been looking at ways to improve scalability, survivability and dynamism of replication technology. Additionally, there are a number of recent application domains that exhibit new and challenging models for information replication. For example, advances in storage technology permit processes to share information by directly accessing data on disks that are connected to a storage area network (SAN), thereby avoiding going through a file system service. This form of direct data sharing necessitates coordination among processes contending for access to data, and presents new building blocks for doing it. New needs are also re-shaped by novel services such as Jini, a global resource discovery and location tool that allows anonymous and transient clients to be serviced; by Java-spaces, a universal shared data space; by Oceanstore, an eternal storage archive that is built of peers that have an economical incentive to cooperate; by Publius, an anonymous and survivable publishing archive; and others. Many other peer-to-peer (P2P) systems offer the potential of a truly survivable settings, but on the other hand, pose challenges of scale, dynamism and trust issues.