Cell architecture

This document proposes a cell architecture for large scale services. It begins by discussing basic architectures and their overhead issues. It then presents a query-off architecture that separates reads and writes across master and slave servers. This helps address workloads where reads outnumber writes. The document then introduces cell architecture, where users are partitioned into cells that each have their own full-set of servers to handle all requests. This allows scaling while limiting failures to specific cells. Benefits include easy extension and isolation of failures, while liabilities include needing more servers to build full cells.