How to avoid marshaling when using shared memory

1 post / 0 new
Yuya MARUYAMA's picture
Last seen: 7 years 6 months ago
Joined: 11/10/2015
Posts: 1
How to avoid marshaling when using shared memory


I'm using Connext DDS and evaluating its trasnport time in different QoS cases for robotics development.

In my evaluation, transport by shared memory is "a littele" faster than transport by local loop back. It was not as fast as I thought at first. Generally, shared memory communication does not require serialization / deserialization, so I expected that it is much faster than the other. 

After research, I heared that Connext provide the option to use shared memory to communicate between endpoints which are on the same system but that still requires serialization / deserialization. I recognized that Connext DDS still marshals messages for various tools (logger, observer, etc.) even when using shared memory.

I woudld like to optimize shared-memory-transport at the expense of versatility. Is this marshaling inevitable by DDS configuratoin? 



Best regards,