What is Data Replication?

What is Data Replication?

Data Replication is the process of continuously copying data from one database or storage node and keeping copies on other servers, often in different locations. In Data Lakehouses and mission-critical Relational Databases, Data Replication is not used only to “back up” data; it is a prerequisite for High Availability, better Read Performance, and surviving hardware or site failures.

If a global bank stores its entire transactional ledger on a single Oracle server in New York, that server is a single point of failure. If the power grid in New York fails, or a disk fails, the bank cannot process transactions. Data Replication addresses this by copying every transaction executed in New York to a standby server in London and a third server in Tokyo.

The Architecture of Synchronization

Data Replication is implemented through low-level database synchronization mechanisms, broadly categorized by how they trade latency for durability.

1. Synchronous Replication (No Data Loss on Failover)

This is the strictest form of replication, used for mission-critical financial transactions. When a user deposits $500 into the primary New York server, the New York server does not simply confirm the deposit. It first sends the $500 transaction across the ocean to the London replica server. The New York server waits until the London server confirms it has successfully written the data to its own hard drive. Only then does the New York server tell the user “Success.”

• Pro: A committed transaction is not lost if the primary fails. If New York goes offline a millisecond later, London has the data.
• Con: It is slow over long distances. Each commit waits at least one network round trip, and light in fiber optic cables crossing the Atlantic takes tens of milliseconds.

2. Asynchronous Replication (High Performance)

This is common for large global web applications and NoSQL databases (like Apache Cassandra) where speed matters more than strict consistency. When the user updates their profile picture in New York, the New York server tells the user “Success!” right away and moves on to the next task. In the background, it streams the update to London.

• Pro: Faster writes; the user does not wait for transatlantic network latency.
• Con: Eventual Consistency. If the New York server fails a few milliseconds after the user uploads the picture, the update can be lost because it had not synced to London yet.

The Master-Slave and Multi-Master Topologies

Data engineers architect replication networks based on the specific traffic needs of the application.

• Master-Slave (Active-Passive): Only the Master node (New York) accepts WRITE commands (new data). The Slave nodes (London, Tokyo) are Read-Only. If millions of users simply want to read data, the global Load Balancer routes the read traffic to the Slaves, reducing the CPU load on the primary Master server.
• Multi-Master (Active-Active): More complex. Both New York and London accept new WRITE commands at the same time. The database engines need a conflict resolution strategy (such as last-writer-wins timestamps or vector clocks) so that if two users edit the same row at the same time on two different continents, the copies converge to a consistent result.

Summary of Technical Value

Data Replication is a core mechanism for global scale and resilience. By synchronizing data across geographically separated server clusters, Data Replication supports High Availability, speeds up global reads by routing users to a nearby server, and gives organizations a way to survive hardware and data center failures.

Learn More

To learn more about the Data Lakehouse, read The Lakehouse Built for Everyone by Alex Merced. You can find this and other books by Alex Merced at books.alexmerced.com.