Traditional approaches to managing data have relied on a centralized architecture – using either a central database to store information, or requiring a master-slave architecture with a central master server to manage the system. These approaches do not scale to address the modern scale of data because they have a central bottleneck that limits scaling. A scale-out architecture delivers unprecedented scale because it has no central bottlenecks. Instead, multiple servers work together as a grid without any central database or master and more servers can be added or removed on-demand.
Scale-out grid architectures are harder to build because they need to be designed from the ground up to not only distribute the workload across a set of processes but also need to provide fault-tolerance so if any of the processes fails the overall system is not impaired.