Ben,

> While allowing 64 byte messages  would be faster for messages from size
> 30-64 am I right in saying it would also incur a penalty for all messages of
> size greater than 64 bytes ?  Since the 64 byte data would be allocated and
> at the least would incur a memory penalty.

Yes, number of bytes added to MAX_VSM_SIZE must be added to allocated 
memory for *each* message.

> Also I suspect the transport
> affects the results eg for Core2 up to 128 bytes copying was faster for
> inproc

It's definitely possible. It depends on cache-line size and many other 
factors. In the particular case of inproc transport on a SMP system, 
copying the data and moving them to a different core using a single 
memory barrier may turn out to be faster than passing pointers and then 
forcing the CPUs to share data via memory coherency subsystem, i.e. on 
cacheline-by-cacheline basis.

Martin
_______________________________________________
zeromq-dev mailing list
[email protected]
http://lists.zeromq.org/mailman/listinfo/zeromq-dev

Reply via email to