IPv6 network boundaries vs. IPv4
Is anyone out there setting up routing boundaries differently for IPv4 and IPv6? I'm setting up a network where it seems to make sense to route IPv4, while bridging IPv6 -- but I can be talked out of it rather easily. Years ago, I worked on a academic network where we had a mix of IPX, DECnet, Appletalk, and IP(v4). Not all of the routers actually routed each protocol -- DECnet wasn't routable, and I recall some routers that routed IPX, while bridging IP... This all made sense at the time -- there were IPX networks that needed to be split, while IP didn't need to be. DECnet was... DECnet -- and Appletalk was chatty, but useful. I keep hearing the mantra in my head of: "I want my routers to route, and my switches to switch." I agree wholeheartedly if there is only one protocol -- but with the mix of IPv4 and IPv6, are there any folks doing things differently? With a new protocol in the mix are the lessons of the last 10 (or so) years not as clear-cut?
John Osmon wrote:
Is anyone out there setting up routing boundaries differently for IPv4 and IPv6? I'm setting up a network where it seems to make sense to route IPv4, while bridging IPv6 -- but I can be talked out of it rather easily.
Years ago, I worked on a academic network where we had a mix of IPX, DECnet, Appletalk, and IP(v4). Not all of the routers actually routed each protocol -- DECnet wasn't routable, and I recall some routers that routed IPX, while bridging IP...
This all made sense at the time -- there were IPX networks that needed to be split, while IP didn't need to be. DECnet was... DECnet -- and Appletalk was chatty, but useful.
I keep hearing the mantra in my head of: "I want my routers to route, and my switches to switch." I agree wholeheartedly if there is only one protocol -- but with the mix of IPv4 and IPv6, are there any folks doing things differently? With a new protocol in the mix are the lessons of the last 10 (or so) years not as clear-cut?
Hi John, I remember old DECNET, DDCMP, IPX and NetBios days. I used to have a couple of 19.2 kilobaud async lines, 2 arcnets and an ethernet (thinwire technology but on RG13U cables, almost yellow wire and UHF connectors - PL-259 like CB-radio). DDCMP could route, IPX could and NetBios was riding on either IPX or DDCMP so it did not matter. Later the DDCMP async was replaced with a lots of switches and repeaters. Whe used to have a backbone (yellow cable) connecting two VAXes and a repeater that was feeding some 8 thinwires. Half of the thinwires were feeding DECNET Terminalservers and PCs the other half were IPX with a single one Netware server and lots of PCs. In its best times the network was seeing some 1000 hosts. Everything was running 10 MBit ethernet. there were 9 segments and no routers. I have seen you could put some 30 NetBios PCs into a single segment or more than 200 DECNET hosts if they were connected via switches and thinwire transceivers. Today without thinwire or yellow cable and with switches that can do 1 Gbit between switches and 100 Mbit to devices you should be able to keep some 1000 hosts within a single switched network. NAT-routers seem to have a limit of some 250 hosts within a single 255.255.255.0 network. I dont know if those boxes really can do 250 or if their MAC address tables break even earlier. I have seen those boxes missbehave when a bad ethernet adapter randomly changed its MAC address. There are quite some link local things in IPv6 so it makes a lot of sense to keep them within a single network - beside that nasty /64 habit that suggests forget radvd and automatic addresses but have an IPv4 address of the 192.168... variety and use 6to4 adressing for your local network. I was running my own network, 4 IPv4 networks and 3 IPv6 networks without routers, only switches :) the 6to4 trick helped me survive but now I dont know if the IPv6 boxes were really seeing each other other simply using 6to4 routes :) Kind regards Peter and Karin -- Peter and Karin Dambier Cesidian Root - Radice Cesidiana Rimbacher Strasse 16 D-69509 Moerlenbach-Bonsweiher +49(6209)795-816 (Telekom) +49(6252)750-308 (VoIP: sipgate.de) mail: peter@peter-dambier.de mail: peter@echnaton.arl.pirates http://iason.site.voila.fr/ https://sourceforge.net/projects/iason/ http://www.cesidianroot.com/
Just for the record about DECNet: At the peak of population, I managed naming and addressing assignments for a DECNet network with just over 8000 nodes. Local routers were mostly Digital Equipment, some wide area used Cisco. After a major split in the network, the remaining 3500 or so Phase IV nodes coexisted happily with AppleTalk, IPX, and IP hosts on a Cisco backbone. My multiprotocol workstation was an Apple Macintosh IIci. Of course, by now the routing network is all IP based with several tens of thousands of routes in the default-free internal network. Cutler When I switch, it is from Windows to Mac OS X At 8/26/2007 11:06 AM +0200, Peter Dambier wrote: John Osmon wrote:
I<snip> Years ago, I worked on a academic network where we had a mix of IPX, DECnet, Appletalk, and IP(v4). Not all of the routers actually routed each protocol -- DECnet wasn't routable, and I recall some routers that routed IPX, while bridging IP... <Snip> I remember old DECNET, DDCMP, IPX and NetBios days. I used to have a couple of 19.2 kilobaud async lines, 2 arcnets and an ethernet (thinwire technology but on RG13U cables, almost yellow wire and UHF connectors - PL-259 like CB-radio).
DDCMP could route, IPX could and NetBios was riding on either IPX or DDCMP so it did not matter. <snip/> In its best times the network was seeing some 1000 hosts. Everything was running 10 MBit ethernet. there were 9 segments and no routers. I have seen you could put some 30 NetBios PCs into a single segment or more than 200 DECNET hosts if they were connected via switches and thinwire transceivers. Today without thinwire or yellow cable and with switches that can do 1 Gbit between switches and 100 Mbit to devices you should be able to keep some 1000 hosts within a single switched network. <snip/> -- Peter and Karin Dambier Cesidian Root - Radice Cesidiana Rimbacher Strasse 16 D-69509 Moerlenbach-Bonsweiher +49(6209)795-816 (Telekom) +49(6252)750-308 (VoIP: sipgate.de) mail: peter@peter-dambier.de mail: peter@echnaton.arl.pirates http://iason.site.voila.fr/ https://sourceforge.net/projects/iason/ http://www.cesidianroot.com/ - James R. Cutler james.cutler@consultant.com
Date: Sat, 25 Aug 2007 23:56:29 -0600 From: John Osmon <josmon@rigozsaurus.com> Sender: owner-nanog@merit.edu
Is anyone out there setting up routing boundaries differently for IPv4 and IPv6? I'm setting up a network where it seems to make sense to route IPv4, while bridging IPv6 -- but I can be talked out of it rather easily.
Years ago, I worked on a academic network where we had a mix of IPX, DECnet, Appletalk, and IP(v4). Not all of the routers actually routed each protocol -- DECnet wasn't routable, and I recall some routers that routed IPX, while bridging IP...
DECnet not routable? Not even close to true. At one time DECnet was technically well ahead of IP networking and far more commonly used. It was not until about 1993 that IP traffic passed DECnet as the dominate protocol and ESnet continued to route DECnet, mostly to support the High Energy Physics community. When the Hinsdale fire segmented tie IP Internet in 1988, the global DECnet Internet survived, albeit with limits bandwidth between the coasts. DECnet was far from perfect and, over time, IP surpassed it in terms of both performance and robustness, but it was not only routable, but globally routed long ago.
This all made sense at the time -- there were IPX networks that needed to be split, while IP didn't need to be. DECnet was... DECnet -- and Appletalk was chatty, but useful.
AppleTalk was a royal pain! Gator boxes and FastPaths would go insane and saturate the network with broadcasts. But AppleTalk did have some really neat features.
I keep hearing the mantra in my head of: "I want my routers to route, and my switches to switch." I agree wholeheartedly if there is only one protocol -- but with the mix of IPv4 and IPv6, are there any folks doing things differently? With a new protocol in the mix are the lessons of the last 10 (or so) years not as clear-cut?
Most routers are a blend of router and switch. The Cisco 6500 and 7600 boxes are probably the most popular large router in the world, but the heart of each is a Catalyst switch. So, the switch switches and the router routes, but they are both the same box. At a major networking show we would switch the IPv6 back to the core routers because of bugs in the IPv6 implementations on many systems. You do what works best for your network. If it means switching IPv6, so be it. This is probably especially true when the router is from a company that charges substantially extra for IPv6 software licenses. If the is only limited IPv6 traffic, switching to a central router might not only be technically the best solution, but the most reasonable fiscal approach. -- R. Kevin Oberman, Network Engineer Energy Sciences Network (ESnet) Ernest O. Lawrence Berkeley National Laboratory (Berkeley Lab) E-mail: oberman@es.net Phone: +1 510 486-8634 Key fingerprint:059B 2DDF 031C 9BA3 14A4 EADA 927D EBB3 987B 3751
OT: He probably meant MOP and LAT are not routable, man that brings back memories. Kevin Oberman wrote:
Date: Sat, 25 Aug 2007 23:56:29 -0600 From: John Osmon <josmon@rigozsaurus.com> Sender: owner-nanog@merit.edu
Is anyone out there setting up routing boundaries differently for IPv4 and IPv6? I'm setting up a network where it seems to make sense to route IPv4, while bridging IPv6 -- but I can be talked out of it rather easily.
Years ago, I worked on a academic network where we had a mix of IPX, DECnet, Appletalk, and IP(v4). Not all of the routers actually routed each protocol -- DECnet wasn't routable, and I recall some routers that routed IPX, while bridging IP...
DECnet not routable? Not even close to true. At one time DECnet was technically well ahead of IP networking and far more commonly used. It was not until about 1993 that IP traffic passed DECnet as the dominate protocol and ESnet continued to route DECnet, mostly to support the High Energy Physics community. When the Hinsdale fire segmented tie IP Internet in 1988, the global DECnet Internet survived, albeit with limits bandwidth between the coasts.
DECnet was far from perfect and, over time, IP surpassed it in terms of both performance and robustness, but it was not only routable, but globally routed long ago.
This all made sense at the time -- there were IPX networks that needed to be split, while IP didn't need to be. DECnet was... DECnet -- and Appletalk was chatty, but useful.
AppleTalk was a royal pain! Gator boxes and FastPaths would go insane and saturate the network with broadcasts. But AppleTalk did have some really neat features.
I keep hearing the mantra in my head of: "I want my routers to route, and my switches to switch." I agree wholeheartedly if there is only one protocol -- but with the mix of IPv4 and IPv6, are there any folks doing things differently? With a new protocol in the mix are the lessons of the last 10 (or so) years not as clear-cut?
Most routers are a blend of router and switch. The Cisco 6500 and 7600 boxes are probably the most popular large router in the world, but the heart of each is a Catalyst switch. So, the switch switches and the router routes, but they are both the same box.
At a major networking show we would switch the IPv6 back to the core routers because of bugs in the IPv6 implementations on many systems.
You do what works best for your network. If it means switching IPv6, so be it. This is probably especially true when the router is from a company that charges substantially extra for IPv6 software licenses. If the is only limited IPv6 traffic, switching to a central router might not only be technically the best solution, but the most reasonable fiscal approach.
On Mon, Aug 27, 2007 at 07:12:54AM -0400, Jason LeBlanc wrote:
OT: He probably meant MOP and LAT are not routable, man that brings back memories.
Yeah, I realy did, but my fingers typed 'decnet isn't routable' because that how the folks I worked with at the time described the issue. I was young at the time, and didn't understand the nuances (as opposed to being older and missing nuances now). My old nuerons took over when I composed the message, sorry for the confusion. Thanks to all the folks that have replied off-list. The (on topic) answers are coming where I expected them: - keep routing boundaries congruent - at local edges / stubs do whatever you want, but do it in private, and wash your hands afterwards (with appologies to R.A. Heinlein)
On 26-aug-2007, at 7:56, John Osmon wrote:
Is anyone out there setting up routing boundaries differently for IPv4 and IPv6? I'm setting up a network where it seems to make sense to route IPv4, while bridging IPv6 -- but I can be talked out of it rather easily.
Why would you want to do that? I've been tempted to do it the other way around, though. In a hosting environment, you can end up with a bunch of /24s dumped on a broadcast domain with a number of different customers but the addresses so intermingled that you can't give each customer their own VLAN. With IPv6, there is enough address space to give each customer a VLAN and and address block to go along with that, which is a lot cleaner.
On Sat, 25 Aug 2007 23:56:29 MDT, John Osmon said:
Is anyone out there setting up routing boundaries differently for IPv4 and IPv6? I'm setting up a network where it seems to make sense to route IPv4, while bridging IPv6 -- but I can be talked out of it rather easily.
We decided to map our IPv6 subnets one-to-one to our IPv4, so each of our routed /22 to /27 subnets gets a /64 IPv6 prefix. This however was just due to the fact that our topology permitted that - your mileage may vary.
participants (7)
-
Iljitsch van Beijnum
-
James R. Cutler
-
Jason LeBlanc
-
John Osmon
-
Kevin Oberman
-
Peter Dambier
-
Valdis.Kletnieks@vt.edu