A recent posting on comp.dcom.cisco suggested that you can't have a 7-bit subnet. For example, 192.239.0.0 255.255.254.0 should be equal to two class C networks, but the two "subnets" would be the all zeros and the all ones subnets. Apparently this goes against the Host Requirments RFC. Would this be a vaild CIDR route? Thanks.... Eric, As I replied privately, I believe this will work with CIDR "supernetting" because no "all-nets" broadcast addresses are defined. This issue may be more sticky as we move toward a truely classless addressing plan, though - consider a CIDR block which is carved out of a "class A" network, for example... Anyone else care to comment? --Vince
I strongly agree. The issue of an "all-nets" broadcast is inconsistent with classless network routing. We should be taking steps to eliminate knowledge of class in all router and host implementations, as flexibility here will certainly be helpful in increasing the lifetime of the IPv4 protocol suite, and transitions from that suite. I always thought the "all nets" broadcast was a broken idea for many reasons, this being one of them. OSPF was designed to be classless internally, but have defaults that are somewhat based on class boundaries. In general, it should be possible emulate "class" for human factors reasons if you implement everything in a truly classless way. Other routing protocols are making this transition as well. It's sort of back to the future, since I believe the original NSF backbone (56 Kbps fuzzballs which did a pure mask and match forwarding decision) had this capability! Thanks, Milo
Milo, Since you raised the point about the Fuzzballs, I would also like to point out that the Merit PDP-11 operating system has had CIDR routing capabilities since 1985 or so. Obviously the wave of the future. On a more substantial note, one issue that was raised at the IETF concerned the idea of subnetting with CIDR (as opposed to supernetting). When do you think that the NIC(s) will be able to hand out pieces of what we now think of as class A nets, for example? My thought is that not only will a very large portion of the Internet need to be CIDR-ized before this happens, but several routers will need to have significant changes to the way forwarding works. Does anyone agree with this? Mark
From: "Milo S. Medin" (NASA ARC NSI Office) <medin@nsipo.nasa.gov> To: Vince Fuller <vaf@Valinor.Stanford.EDU> CC: Erik Sherk <sherk@sura.net>, regional-techs@merit.edu
I strongly agree. The issue of an "all-nets" broadcast is inconsistent with classless network routing. We should be taking steps to eliminate knowledge of class in all router and host implementations, as flexibility here will certainly be helpful in increasing the lifetime of the IPv4 protocol suite, and transitions from that suite. I always thought the "all nets" broadcast was a broken idea for many reasons, this being one of them.
OSPF was designed to be classless internally, but have defaults that are somewhat based on class boundaries. In general, it should be possible emulate "class" for human factors reasons if you implement everything in a truly classless way. Other routing protocols are making this transition as well. It's sort of back to the future, since I believe the original NSF backbone (56 Kbps fuzzballs which did a pure mask and match forwarding decision) had this capability!
Thanks, Milo
Well, I stand corrected! In many ways, the mask and match approach is a very natural way to think about things, but many of us who were raised on BSD unix had class drilled into the way we used those systems as routers and hosts. I think it's sort of like learning BASIC as your first computer language. You're hobbled for life! :-) You raise a good point about NIC's handing out such addresses. This will be problemmatic, but straightforward engineering. Enough to employ some database types for awhile. The biggest problem I don't think is with the routers either. I think it's with the way the in-addr.arpa kludge in the DNS will deal with address allocation on non-8bit boundaries. This is a mess. Automation can help, but that's going to be ugly. We should try and allocate chunks of address space on 8 bit boundaries until we can field a more flexible scheme... Most of the major router vendor's CURRENT releases support classless IP forwarding operations, I think mostly as a consequence of Variable Length Subnet Mask support required in OSPF. Since OSPF is pretty much considered a must-have in the router game, people had to invest in the labor to fix this up. It took awhile, but by and large we are there, though many of the Unix vendors who use these systems as routers still haven't taken the code from CSRG and LBL that would make this work for them. But I think the OSPF push for VLSM will make life a lot easier on the BGP4 folks, since they won't have to coerce the vendors for classless forwarding support. Thanks, Milo PS You also made me think that we should probably try and get the InterNIC folks to start working on making their address allocation and database stuff support "classless" allocation. I'll bring this up with NSF next time I'm in DC...
On a more substantial note, one issue that was raised at the IETF concerned the idea of subnetting with CIDR (as opposed to supernetting). When do you think that the NIC(s) will be able to hand out pieces of what we now think of as class A nets, for example? My thought is that not only will a very large portion of the Internet need to be CIDR-ized before this happens, but several routers will need to have significant changes to the way forwarding works. Does anyone agree with this? Mark
Here's my two cents. Its way too early to be considering handing out subnets of class A's. For class A's that want be announced as peices of class A, that should be fine as long as they have a single provider who is CIDR/BGP4 capable and can aggregate to a full class A for the rest of the world. The decision gets a bit more difficult for a class A that wnats to be subnetted and has multiple providers that are all CIDR/BGP4 capable. I'd say that this is still possible since the multiple providers can each aggregate to a full class A and pass the aggregate to the rest of the world, who in turn can do the primary, secondary, etc thing pointing at the whole class A. I don't think there is a case where you would be worse off than when announcing the whole class A to each CIDR/BGP4 provider. This sounds to me like it would cover what the class A networks would like to do. Did I miss anything? I wasn't at Amsterdam or the regional techs meetings, so I extend my appologies if this is meaningless or wrong or previously discussed. Curtis BTW - Are the routers needing significant changes ours? If so, we're not very far from deploying gated and fixing this.
participants (4)
-
Curtis Villamizar
-
Mark Knopper
-
Milo S. Medin
-
Vince Fuller