Google’s Urs Hoelzle, Vice President of Operations and Vice President of Engineering, provided a rare insight into behind-the-scene architecture of Google when spoke at EclipseCon 2005.

The key to the speed and reliability of Google search is cutting up data into chunks and replicating the data across multiple cheap machines that are placed in interconnected nodes to ensure plenty of redundancy.

To achieve high performance and high availability of Google search, Google uses three software systems that are in-house built to route queries, balance server loads and make programming easier.

Google File System replicates the data to ensure redundancy and failed machine will not cause any disruption, Map/Reduce Framework provides automatic and efficient parallelization and distribution while Global Work Queue schedules queries into batch jobs and places them on pools of machines.

Full Article