Date: Tue, 12 Oct 1999 02:17:44 -0400 (EDT) From: Mike Nowlin <mike@argos.org> To: Mark Summerfield <m.summerfield@ee.mu.oz.au> Cc: Archie Cobbs <archie@whistle.com>, Garrett Wollman <wollman@khavrinen.lcs.mit.edu>, freebsd-net@FreeBSD.ORG Subject: Re: arp errors on machines with two interfaces Message-ID: <Pine.LNX.4.05.9910120139380.15101-100000@jason.argos.org> In-Reply-To: <199910110153.LAA19461@mullian.ee.mu.OZ.AU>
next in thread | previous in thread | raw e-mail | index | archive | help
> At 17:23 10/10/99 -0700, Archie Cobbs wrote: > >Garret, I'd be interested in seeing what references you have that > >say having two NICs on the same wire (with NON-overlapping net ranges) > >is somehow broken or illegal or technically incorrect. > > > >I had always assumed this was perfectly legal, just like running > >AppleTalk and TCP/IP on the same wire is perfectly legal. > I ran into this problem (actually, I'm still dealing with it) a few years ago when we got a dedicated Internet line at work... When the network was first set up, we used 192.168.2.0/24 addresses for everything. With the new IP block from the ISP, we then had 208.132.36.128/25, which I started working into the network... Due to some problems with our application software (hard-coded IP addresses that would have cost $1000's to change), we couldn't switch about 50% of the equipment to the new IP block... Turns out that some of the machines had two IP addresses assigned to them, and some of those ended up getting a second ethernet card - one for 192, one for 208. Aside from a few headaches caused by having to manually do the routing tables on all the machines, everything was OK until we started putting some WAN routing equipment in that used RIP with routing info that changed every time one of our remote LANs connected -- life suddenly got complicated. The routing tables would suddenly clear out, then build back up again, then empty out again. Sometimes they would build up correct info, sometimes not... An address/netmask combination basically refers to the addresses that a certain machine can talk directly to in a single hop. When you start mixing different combinations on a shared ethernet, RIP (and probably other routing protocols) get really confused -- they hear two different broadcasts that are essentially contradicting themseleves. Your results will vary depending on what version/release of the protocol you're running on -- our DEC UNIX boxes ended up setting the ethernet cards to something like 192.168.2.4/0, which made them completely lose track of the 208 address, and the default gateway, and the Linux boxes did something similar. The FreeBSD boxes started reporting messages like: arp: 208.132.36.174 moved from 00:00:c0:0b:01:81 to 00:40:05:1e:df:19 on ed1 arp: 208.132.36.174 moved from 00:40:05:1e:df:19 to 00:00:c0:0b:01:81 on ed1 arplookup 192.168.2.15 failed: host is not on local network (just happened to have that saved).... I finally figured out a way to make everything work, as long as our ISP line didn't go down - the DNS box started throwing fits if that happened..... With the installation of some new systems (Y2K fixes), I had the opportunity to fix a lot of these problems.... I "worked" a couple 40- and 24-port HP intelligent switches into the purchase orders, along with some new wiring. Between the VLAN capability of the switches and the extra wiring, we've fixed about 95% of the problems. Things are pretty much back to normal as far as the standards go, and none of the 192 (or the new 10.x.0.0/16 addresses) can hear the 208 machines, and RIP is now working fine. The only reason it's working now is that there are logically three ethernets - 208, 192, and 10 (which goes to another router to break it down into 10.0.0.0/24 chunks). Long winded, I know -- but you wanted an example... --mike To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-net" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.LNX.4.05.9910120139380.15101-100000>