A better way to spend our resources and to gain *real* confidence with the newly deployed CIDR is to follow the strategy used by Havard:
>1. Announce a CIDR aggregate >2. Remove individual components of the aggregate and >test its impact on the connectivity. Are you advocating to use real operational nets as guinea pigs :-)? I feel, at least in this stage which we know the exsitence of CIDR-traffic-black-holing-ASs, Eric-jan's approach is a more appropraite one. --Jessica
Jessica, I suspect that most providers have a classful network number within a CIDR block that does not have production hosts on it which they can use for this kind of testing. Thus, this does not have to be a test using live traffic. Testing would look like: Send in specific classful routes including the "test network" Send in aggregate route Test connectivity ala Havard ... Withdraw the test network route Test connectivity ala Havard from a test station on the "test network" If everything is okay, then removing the rest of the specific network number routes would seem to be okay. It may be necessary to this on an aggregate by aggregate basis until all of the operational configuration issues are wrinkled out. This would seem to meet your safety concerns. peter
Peter,
I suspect that most providers have a classful network number within a CIDR block that does not have production hosts on it which they can use for this kind of testing.
I think so.
Thus, this does not have to be a test using live traffic. Testing would look like:
>1. Send in specific classful routes including the "test network" >2. Send in aggregate route >3. Test connectivity ala Havard ... >4. Withdraw the test network route >5. Test connectivity ala Havard from a test station >on the "test network" >6. If everything is okay, then removing the rest of the specific network number routes would seem to be okay. I think this is a good approach and it is basically what Eric-jan is doing. He did not do Step1 and step4 which are not necessary. It achives the goal of detecting existing CIDR-traffic-black-holing-ASs and moving towards the withdrawal of the living classful routes included in this aggregate.
It may be necessary to this on an aggregate by aggregate basis until all of the operational configuration issues are wrinkled out.
This would seem to meet your safety concerns.
Yes. Again, I think this approach be used only for this stage when we can name the ASs which are neigher do CIDR nor defaulting yet. I still think this is a better approach than using live nets to verify the brokeness of connectivity to CIDR routes. --Jessica
What ASs do we know about that do not do BGP4 and do not have a default? The only ones that I have heard about are the CIX and NSI. Are there any others? [My understanding is that NSI has some workarounds in place and is fairly close to BGP4.] --asp@uunet.uu.net (Andrew Partan)
participants (3)
-
asp＠uunet.uu.net -
Jessica Yu -
Peter S. Ford