Moreover so as to parallelize the switch even when blocking I/O is used, the code tries to write down a small quantity of data to every fd in a loop, so that the kernel will ship the packets within the background to multiple slaves at the same time. http://Mycrazymail.net/__media__/js/netsoltrademark.php?d=products-dd.com%2Fcategory%2Fpocket-wifi%2F