The Backplane database is a replicated, transactional, fault-tolerant relational database core implementing a subset of SQL. Our goal is to develop the database into a fully SQL-compliant system. The main feature of the Backplane database is its peer-to-peer replication technology, which allows essentially infinite scaleability for read-only transactions and moderately scaleable read-write transactions which require only a quorum of the replication peers to be up in order to be fully operational. The database guarantees transactional consistency across the entire replication group. The replication technology is extremely fault-tolerant and is able to deal with individual peers going up and down, even in the middle of a query. This allows you, for example, to implement a farm of web servers talking to a farm of database servers and maintain transactional consistency across CGI executions running on different web servers. A single user session can in fact jump between web servers on every submit and still maintain complete consistency.

The replicator is capable of managing many independent databases. You have complete control over which replication hosts replicate which database(s), and complete control over which replication hosts simply snapshot a database (snapshots do not take part in the quorum-based commit protocol but can take part in read-only queries. Snapshots are typically used for batch or near-realtime backups).

It is especially important to note that the Backplane replication system uses a quorum-based two-phase commit protocol. This means that if you have 5 PEERs replicating a database (and any number of additional SNAPshot hosts), only 3 of those peers need to be up in order for the database to be fully operational. Most other databases, include many major commercial databases, implement only fully-synchronous replication or asynchronous-push technologies. Fully-synchronous technologies are not fault-tolerant - one downed host takes down the whole system. Asynchronous push technologies cannot guarantee transactional consistency between replication hosts and require a significant (sometimes immense) effort in building conflict resolution procedures. The Backplane replication technology only requires a quorum of PEERs to be operational and implements background synchronization to bring hosts that lose connectivity up to date. The Backplane replication technology is fully transactionally coherent between replication PEERs whether or not they are fully synchronized, which makes it ideal for web farms. Additionally, many databases cannot replicate meta-SQL commands such as alter table reliably (or at all). The Backplane replication technology replicates tables natively at the row-level and you can issue every supported SQL command as easily in a replicated environment as you would in a single-host environment.