| layout | title |
|---|---|
post |
Overview |
The "secret sauce" of Syndicate is its ability to be adapted to handle any application's storage semantics with minimal changes. It does so by processing reads and writes as a sequence of "pipelines" that run through a set of network-distributed "gateway" processes.
A "gateway" is a Syndicate process that handles reads and writes from applications and other gateways. An application reads and writes data through a co-located gateway. That gateway talks to other gateways in order to carry out the request according to the application's expected storage semantics.
For example, Alice may run a gateway locally that implements a FUSE filesystem, and Bob may run a gateway remotely that can load and store data to and from Amazon S3. Alice's application writes a lot of redundant data, so she programs her gateway to deduplicate writes before making them persistent.
When Alice's application writes data, her gateway checks to see if the data has never been written before. If so, it sends it to Bob's gateway, which will then send them to S3 for storage. Later, when Alice's application reads data, her gateway will pull data from Bob's gateway (which in turn fetches it from S3).
The deduplication logic in Alice's gateway and the S3 interfacing logic in Bob's gateway are both examples of gateway drivers. Gateways are undifferentiated processes; their application-specific functionality comes from the drivers they run.
Gateway Types
Data Model
Data Transport
Data Indexing
Data Security
Deployment
Drivers
Data Serialization
Caching