Date: Sun, 8 Jun 2008 14:09:58 +0200 (CEST) From: Wojciech Puchar <wojtek@wojtek.tensor.gdynia.pl> To: Jon Radel <jon@radel.com> Cc: "Gelsema, P \(Patrick\) - FreeBSD" <freebsd@superhero.nl>, freebsd questions <freebsd-questions@freebsd.org> Subject: Re: NFE setting manually to 1000baseT and half duplex Message-ID: <20080608140911.S40202@wojtek.tensor.gdynia.pl> In-Reply-To: <484BBC23.90505@radel.com> References: <56297.82.95.198.17.1212870050.squirrel@webmail.superhero.nl> <484BBC23.90505@radel.com>
next in thread | previous in thread | raw e-mail | index | archive | help
> Even if you do have hardware that supports half-duplex gigabit ethernet > on both ends, the need to do carrier extension for any frame shorter > than 512 bytes so that CSMA/CD actually works on a reasonable sized > cable, does horrible things to your throughput if you've got lots of > small frames. (In other words, at gigabit speeds, frames smaller than > 512 bytes zip down the wire so quickly that you can no longer reliably > detect collisions, so the frames all get padded.) I'm having trouble wrapping > my head around any circumstances other than horribly, horribly broken > hardware or software where half-duplex would increase your performance over > full-duplex. actually there are no gigabit devices incapable of full-duplex. > ethernet hardware I've ever touched has been incapable of doing > half-duplex when it's being used at gigabit speeds. The specs for doing > it exist more for theoretical completeness than out of practical > utility. See, for example > > http://www.intel.com/network/connectivity/resources/doc_library/white_papers/solutions/copper_guide/gig_over_copper.htm at 10Gbit/s specs dropped half-duplex and collision detection at all.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20080608140911.S40202>