At 2:39 PM +0000 12/19/02, steve wrote: >Howard, > >just my 2 pence > >you know ... > >funny you should say about BGP ... I was just thinking that the other day... > >but I personally don`t agree with the new protocol theory... >I personally don't claim to be able to do this....I can barely plug in a >switch ....but >as far as I am aware ....the ISP BGP world is an all seeing all knowing >world were all OX amount of routes are seen by everyone ..and I believe this >is where the problem lies.
There's a subtle difference, and also some history here. The absolute number of routes isn't the problem. We can build perfectly good memory structures and search algorithms to find 10 million routes. The problem is the rate of change of the routes, constantly being added and withdrawn, which has a couple of significant effects. First, it's a processor load on individual routers. That can be dealt with, I think -- in fact, multiprocessing in the routing control plane is one of my research interests. The big problem comes with global instability. The Internet never really converges as a whole, and the protocol designers in the IRTF and IETF have pretty much agreed to that. > >When designing a routing protocol ,there is a basic problem that all >designer`s face is ....links go up/down ...route`s appear and disappear... >the more routes you have the more the protocol has to do ...regardless of >how you get around this fact with fancy techniques ,there will still be a >scalability problem based around a connectivity problem ,the more routes the >more unstable the less your inclined to scale .... >the protocol`s I think can probably made more efficient ,but it does not >address the real problem , >that is the amount of routes that a being added daily make`s any >computational algorithm`s task very difficult . > >the only way in my humble opinion to make this more stable/scaleable is to >back to the OSPF DESIGN NOT PROTOCOL... > >Regionalise .......create Super AS for various regions i.e US UK JP >AUS...and then Tag all routes coming out .. That, indeed, is the fundamental assumption of CIDR and BGP-4. Unfortunately, we are having great market and perceptual problems in changing. You see, CIDR/BGP-4 assumed that significant aggregation in a hierarchy was possible and desirable. The Internet, in broad terms, would be a pyramid, although there would be multiple major carriers at the peak of the pyramid. EGP assumed only one core. Various people, especially Geoff Huston, have demonstrated that the Internet topology is "flattening" away from the pyramid. The IETF PTOMAINE Working Group web page, under www.ietf.org, is a good source here. The growth in AS is not particularly in carriers, but in multihoming users. The users want protocol-independent address space so they aren't locked into a business relationship with a single carrier, and they want to be able to home to arbitrary carriers at different hierarchical levels. Flattening defeats aggregation. IPv6 has some measures that make it easier to switch carriers, but the multihoming problem is harder in v6 than in v4. There are also user expectations of fine-grained high availability that won't work in a highly aggregated environment. > >OK (in an ideal world) this IS NOT the only way of doing things....link 1 of >8000 goes down ...your advertising all 8000 out of one supernet ... > >But atleast in this case only your "Super ASBR`s" if you like.... would only >need to communicate with eachother ... > >perhaps this is what already happen`s ....but i see that a fundamental shift >in the way we network is required and not necessarily a change in protocol >.... > >many thanks > >(I`ll keep my head down now ...i think...i`m only trying to help !!!) > >Steve Message Posted at: http://www.groupstudy.com/form/read.php?f=7&i=59529&t=59529 -------------------------------------------------- FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
