> I played a bit with it and it seems the R5 limit is at about 2MB. A > little less and it not always the same, which makes me suspect, that > some kind of global buffer or limit is used for all ports. Our > implementation has a 65KB limit per message, which is certainly not > adequate. So I guess that using a BMallocIO is a good solution. I would use a chunk size of 512 or 1024, then. > > BTW, R5's write_port() doesn't even report an error, when the message > is too large. It simply doesn't write the message and nevertheless > returns B_OK. Yeah, IIRC, that was one of the problems that JBQ complained about.