Josh Faust wrote:
> Do you have any timing data of this vs. TCP over loopback?  Since 
> you're still doing a serialize/deserialize I would expect them to be 
> fairly similar.
I have not had a chance to seriously evaluate the performance 
difference. It seems that the system load is marginally smaller for 
small images, but I guess the difference would be more visible with 
bigger data.

I'm really curious if there is an advantage to shared memory, and if so, 
at what object size it starts, or what number of objects. I guess one 
has to implement it to see what it brings...

Using serialise/deserialise is I think the only way to do it without 
specialising to one object type. I don't think these functions have a 
huge cost with boost implementation, but I'd be curious to hear about 
any experience out there.

Best

-- 
Dr. Cedric Pradalier
http://www.asl.ethz.ch/people/cedricp