Replication and Consistency in a Distributed Environment

(PREVIOUS TITLE: REPLICATION AND CONSISTENCY: BEING LAZY HELPS SOMETIMES) The problem of consistent access to replicated data has re-emerged as a challenge in recent years with the advent of distributed data warehouses and data marts at the high end, and distributed data in often-disconnected mobile computers at the low end. The fundamental problem, as identified by [5], is that the standard transactional approach to the propagation of updates to replicas is unstable - deadlocks increase as the cube of the number of network sites and as the fourth power of transaction size. This is particularly problematic with relatively long data-mining queries and with mobile transactions. The former access many data items; while the latter effectively live for a long period of time if the mobile computer is disconnected. Thus, deadlock is no longer a rare event with a negligible effect on performance; instead, it is a barrier to the ability of systems to scale up.