At a high-level, most distributed architectures take a similar approach: Spread partitions over multiple nodes, insert/query to any node, route (sub)requests to the appropriate nodes. The unique parts of Timescale is that, unlike Postgres XL or Citus, our design/optimizations all focus on time-series data.
So, as we've mentioned in some other replies, this means that our partitions/chunks are typically based on two-dimensions (time and a primary partition key), chunks across nodes are automatically created and closed (based both on time intervals and sizes), and there are a bunch of time-based distributed query optimizations and time-series-specific functions that we'll provide out of the box.
So, as we've mentioned in some other replies, this means that our partitions/chunks are typically based on two-dimensions (time and a primary partition key), chunks across nodes are automatically created and closed (based both on time intervals and sizes), and there are a bunch of time-based distributed query optimizations and time-series-specific functions that we'll provide out of the box.
More information in whitepaper: http://www.timescale.com/papers/timescaledb.pdf