On Jun 7, 2009, at 1:11 PM, Niki Kovacs wrote: > Rainer Duffner a écrit : > >> Yup. >> If you have the time, you can experiment with the blocksize and see >> where the throughput is best. >> http://unix.derkeiler.com/Mailing-Lists/FreeBSD/questions/2008-09/msg01375.html > > Interesting thread. Guess I'll give it a few spins with different > blocksizes (20k, 200k, 2M, 20M) and time { } the operation. Just > curious. > > Anyway: thanks! > > Niki I suspect that once you get to some small size multiple of the lower level protocol, larger block sizes probably don't matter too much. 512 byte blocks are small, so there would be considerable overhead. 1MB or larger blocks, you have exceeded the probably block size of the lower level driver or hardware, so then it is just being split up by the driver.