I have an API server (Java-based) that's serving data tables via remote procedure calls. The calls are initiated by a client (C++) that receives the tables, and performs heaving number crunching on that data. The data payload is fairly substantial (some megabytes) and latency is important.
Currently, the tables are transferred via JSON or XML which are excellent formats for structured data but have some pretty excessive overhead for large tabular data. A good chunk of time is spent simply packing and unpacking the data. Is there a standard transfer protocol or format for serving things that look like database tables? (Sequences of rows, where each column is a fixed type (e.g. String, int, float, ...) )
Ideally, on the client side, I would receive the payload directly into a memory buffer and construct an iterator that walks through it rather than having to unpack the data.
