Hello,

We use QNX 6.3.0 on a PPC VME multiprocessors system.Each processor, runs a separate nto kernel,  and opens a dual port 
memory ( local / VME ).  

In this system, the communications between processors, are based on native message passing (i.e the messages are posted 
on the VME bus).

Sometimes the process doing the transfer is preempted during this one ( this is a normal behaviour of our system ). 
Later, when the process get again the cpu,  the transfer resumes at the following memory cell. But time to time, the 
transfer resumes few cells above, and more rarely it restarts from beginning of the message !!

Of course there are never transfer error ( the good datas are always moved to the good addresses) but in case of big 
transfer we have a performance impact. 

The instrumention is based on our VME tracer 

I don't know if this behaviour is due to the nto kernel ( cache side  effect or something else ) or to the hardware ? 
May be the same thing is going on in the system RAM, but I have no hardware tool to observe it. 

Thanks for your help.

Philippe.