How is the underlying CQL native binary protocol implemented?

[joy]

Hi, I am curious about the underlying implementation of the CQL native binary protocol. It is also a RPC protocol, right? But how can it outperform thrift as they are both RPC protocol?
The Cassandra c++ driver has already implemented this native binary protocol, could anyone tell me which part of the source code is the core of the binary protocol?

[Sorin Manolache]

On 2014-12-24 06:24, joy wrote:
> Hi, I am curious about the underlying implementation of the CQL native binary protocol. It is also a RPC protocol, right? But how can it outperform thrift as they are both RPC protocol?
> The Cassandra c++ driver has already implemented this native binary protocol, could anyone tell me which part of the source code is the core of the binary protocol?
>

I suppose it is faster because the native protocol is asynchronous. I.e.
the client may send a 2nd, 3rd etc request before the response of the
1st request arrives. So it supports pipelining. I don't know if it
supports out-of-order responses, i.e. if the response of the kth query
may arrive before the response of the nth query, k > n. I think yes but
I don't know.

Also I have the impression that the native frames are shorter than the
thrift frames that transport the same requests/responses. The native
protocol frame contains meta-data for the _heading_ of the table that is
sent in the response. The thrift frame contains meta-data for each
_cell_ of the table that is sent in the response.

For the implementation, have a look at query_request.hpp, response.hpp,
serialization.hpp, and connection.hpp.

Sorin