Why ScyllaDB is Moving to a New Replication Algorithm: Tablets

ScyllaDB is transitioning from a Vnode-based replication strategy to a new algorithm centered around tablets to enhance data distribution and elasticity across clusters. While Vnodes have traditionally been used to evenly distribute data by randomly selecting token space split points, they often result in imbalances and inefficiencies, particularly for small tables. The tablet-based approach allows for a more dynamic and flexible distribution by assigning fixed-size tablets to serve table fragments, enabling the movement of individual tablets to balance loads. This new method also integrates the Raft consensus algorithm, offering advantages like strong consistency and higher throughput, although it introduces challenges such as potential latency during leader failover. Tablets facilitate more efficient data management, offering easier resharding and cleanup processes, and improve load distribution by enabling finer control over where data resides. As ScyllaDB implements this technology, it aims to address the limitations of Vnode partitioning by providing a more scalable and manageable system.