Design Goals

The objective of Unbase is to create a unified data/application framework, which achieves all of the following goals. Unbase could be loosely conceptualized as a sort of P2P Distributed Object Database, minus the “base”.

  • Minimize latency

    Give users a perceptibly-instant experience whenever possible.

  • Scalability

    The system must gracefully handle millions of participants with mixed read/write loads.

  • Regional autonomy (geographical, planetary)

    No significant fraction of the system should be offlined1 due to network partition. Must handle significant latency between regions (minutes, hours, days), as well as relativistic effects. No, this is not a joke.

  • Strong Consistency Model

    Provide system users with strong guarantees via the Infectious Knowledge consistency model

  • Eliminate the distinction between client and server

    Equality for all participating nodes. Policy and capacity are the only factors which should bias decision-making. The client/server model is flawed. Let browsers and other nodes formerly conceptualized as “clients” talk to each other, and even implement core business logic.

    Verify outcomes as part of the replication process, to ensure that they are authorized.

  • Decouple business logic via triggers

    Authorized parties may advertise actions as “triggers”, such that business logic can be initiated automatically on the basis of record edits. These triggers may be synchronous, or asynchronous, with the latter being strongly preferred for most use cases.

  • Integral push-updates

    Data replication and event notification are the same thing. Out-of-band event propagation is nigh impossible to coordinate with in-band consistency models. (RabbitMQ is not your friend)

  • Integral content-filtered subscriptions

    Create and manage distributed content-matching trees, such that interested parties may subscribe to record edits/additions/removals on the basis of a query expression.

  • Integral audit trail

    Given the mechanics of distributed data replication, significant efficiency can be gained by extending the system to handle audit trails natively, rather than storing audit trails in standard tablespace.



1: Using the term “offline” for conversational understanding. There is no such thing as “Online” or “Offline” except as observed from a specific point in space and using a entityive metric.