Tim Crowell (GTE.NET) wrote-
Howdy folks,
I would like to pose a question to the group about the best way to implement the following;
GTE has a customer who is a content provider that we have allocated a class C out of our CIDR block.
They have subsequently also ordered a second transit service from ISP XYZ.
I would make them renumber with a new class c that was not in your CIDR block. Maybe they could get a class C from the swamp?
Our assumptions are: 1. Customer will obtain an AS number to do BGP with both GTE and XYZ. 2. BGP will be established with both ISPs 3. GTE will announce the class 'C' as both a part of our aggregate CIDR block and as a specific /24
No you would just advertise the aggregate CIDR block, they should advertise the more specific route so when you peer with them their route is propagated with their AS number.
4. XYZ will announce the class 'C' as a /24 only.
XYZ should peer with them just as you have done.
5. Both GTE and XYZ will supply a default route.
If the client is carrying the full routing table should be default free, if they chose not to then they will have to decide on which provider should be the default route. Anyway my thoughts on an entity having multiple ISP's would want to have an autonomous system number routing his class C. I guess you could strip his AS and route is as your own but I dont think his would fall in line with current RFC's. I may be wrong though. Someone from this mailing list I am sure can enlighten us.
Explanation/Questions; 1. Does this AS number have to be an officially registered AS or can it be a reserved number?
If they are advertising the route the rest of the world it would have to be registered.
The thought is that the Class 'C' will be announced by both ISPs and strip the customers AS. The AS would only be used to connect between ISPs. It seems extremely wasteful for every little company that wishes a dual homed network would have to get a registered AS.
They have registered domain names and a 24 bit network? Just kidding. This is beyond my political domain.
2. We first had major heartburn with carving the 'C' out because we just couldn't see having to add 2 additional announcements to the internet routing tables but we have come to the conclusion that there is no other way to do it.
Would not there be just one additional route that is propagated by you?
We assume that we have to announce the /24 in addition to our aggregate otherwise XYZ's more specific announcement of our network would route all traffic through them from the internet.
That is correct
It just seems that if there were a large number of these multi-homed Class 'C's that the internet routing table would be flooded. (Maybe thats a part of the problem.
Now you understand why everyone is using Cisco 7500 series routers with 64 to 128 megs of RAM.
3. As a followup, what would you do if a subnetted class 'C' customer who only requires a dozen or so addresses but orders connections to two ISP's. Do you burn a whole Class 'C' ????
Yeah. I dont think anyone will accept nor propagate a route larger than 24 bits.
4. Is there anyway to accomplish what the customer wants that we haven't considered.
If it was me I would first of all register for a class C network address from the NIC in order to not break up you CIDR block. As I said before stripping their AS and adverstising the route with your origin and another ISP would not be thing to do. If I am wrong somebody please correct me.
5. I understand that we will have to submit the Class C to RADB and create a "hole" in our aggregate to effectively represent the network topology.
It would actually just be a more specific route that would seem seemless to you.
PS. If i'm just being stupid about this feel free to say so. I don't pout too long.
Don't worry, everyone has BGP nightmares. Unfortunately it is one of the nightmares that reoccur every once a while.. Chris MacFarland President CompuNet 214.994.0190 chris@computek.net
In message <199608162135.QAA07812@ns1.computek.net>, "Chris MacFarland" writes:
Tim Crowell (GTE.NET) wrote-
GTE has a customer who is a content provider that we have allocated a class C out of our CIDR block.
They have subsequently also ordered a second transit service from ISP XYZ.
I would make them renumber with a new class c that was not in your CIDR block. Maybe they could get a class C from the swamp?
Are you suggesting that all dual homed networks should be renumbered such that they can't be aggregated and can't be reached from a good part of the Internet. I don't think that is a good idea. Are suggesting punishing a customer for picking up a second provider by giving them an unroutable prefix? I hope not.
Our assumptions are: 1. Customer will obtain an AS number to do BGP with both GTE and XYZ. 2. BGP will be established with both ISPs 3. GTE will announce the class 'C' as both a part of our aggregate CIDR block and as a specific /24
No you would just advertise the aggregate CIDR block, they should advertise the more specific route so when you peer with them their route is propagated with their AS number.
I think it is assumed that GTE would be passing on the route heard from the customer. The difference is they would not be passing on the more specific routes they learned from single homed customers in the same blocks, or customer multihomed to them.
4. XYZ will announce the class 'C' as a /24 only.
XYZ should peer with them just as you have done.
Same comment. Announce does not mean "configure a static route".
5. Both GTE and XYZ will supply a default route.
If the client is carrying the full routing table should be default free, if they chose not to then they will have to decide on which provider should be the default route.
There is no harm having a backup. They should not try to load split across the two default routes unless they want extremely poor TCP performance.
It just seems that if there were a large number of these multi-homed Class 'C's that the internet routing table would be flooded. (Maybe thats a part of the problem.
Now you understand why everyone is using Cisco 7500 series routers with 64 to 128 megs of RAM.
I think the percentage of dual homed /24s is small compared to the number of prefixes that have not been aggregated with no good reason at all. Curtis
participants (2)
-
Chris MacFarland
-
Curtis Villamizar