Re: bge & vlan stranges

From: Terry Lambert <tlambert2_at_mindspring.com>
Date: Sat, 02 Aug 2003 16:54:19 -0700
Boris Kovalenko wrote:
> No, this is test machine, I have installed it two days ago and have
> firewall_type="OPEN" in my settings. So I have not disabled MTU path
> discovery You are speaking of. Nevertheless, what is "substracted from
> available MTU?" Why? The correct way it should work:
> 1500 bytes packet + 14 bytes ethernet header + 4 bytes CRC = 1518 bytes
> is standard ethernet frame and
> 1500 bytes packet + 14 bytes ethernet header + 4 bytes 802.1Q tag + 4
> bytes CRC = 1522 bytes of standard 802.1Q encapsulated frame. All 802.1Q
> realizations I know working the same.

The way it *does* work is:

	MTU = 1518
		- 14 bytes ethernet header
		- 4 bytes CRC
	    = 1500

	MTU = 1518
		- 14 bytes ethernet header
		- 4 bytes 802.1Q tag
		- 4 bytes CRC
	    = 1496

...unless you have some strange card magic that lets it send packets
larger than allowed for by the standard.

BTW: the reason the length is what it is by default is that the
ethernet standardard is a CSMA/CD type transport (designed after
the old "Aloha" radio system), and the length is picked to be as
large as possible for the bandwidth to provide for a medium that
has as few collisions as possible.  Effectively, since people
transmit randomly, with a random back-off on collisions, this is
a hash function with the length chosen to represent a best-guess
effort at an 85% fill rate.

Relying on card magic is bad, unless you can guarantee a completely
homogenous environment (i.e. everyone has the same card magic).

Path MTU discovery is an issue on packets with the DF (Don't Frag.")
bit set, since those packets, if they are sent from a system with a
high MTU to another with the same MTU, but an intermediate MTU is
smaller, get "black holed" (hence the term "black hole routes").

-- Terry
Received on Sat Aug 02 2003 - 14:55:29 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:17 UTC