On Wed, Sep 11, 2002 at 09:16:38PM -0400, Joe Abley said:
On Wednesday, September 11, 2002, at 07:39 PM, Simon Blake wrote:
+ having a route propagation path which is different to the packet forwarding path, which is a general problem of route servers on non-trivial layer-2 exchange fabrics;
Sure, but that's an equally good arguement for fixing the exchange fabric, since if you're under some kind of partial reachability cloud, then you're not getting full value for your connection anyway.
Right.
The issue arises when there is a functional layer-2 path from the route-server to your router, but no functional layer-2 path from your router to the next-hop address for particular routes you learn from the route server.
In those circumstances you will black-hole traffic across the exchange instead of losing the route and sending the traffic by some other path (which is what would happen if you only learnt routes directly from the other operator's router, with no route server involved).
It's a failure mode thing, not a value for money thing.
Well, that's debatable - it's still broken, you're not getting value for money, and if the L2 connection over the exchange between the two ISP's is the only path between them, then you're still screwed. I think my observation still stands - if your L2 infrastructure is broken such that you don't have reachability between points A and B on that infrastructure, then that is the real problem, and that's what has to be fixed. What we're really discussing is whether the advantages the route servers bring in normal operation is outweighed by the disadvantages when the L2 exchange goes pear shaped. I think the advantages are greater (because experience tells me that the NZ L2 exchanges are reliable), you think the disadvantages are greater (presumably coz you've seen exchanges elsewhere go mental). I don't think you can broad brush the arguement either way.
There is no contract between ISP A and ISP B who receive each others' routes from the route server, though. That's what I was getting at. That's important to some people, and not just for lawyers-run-the-company, my-hair-is-pointy reasons.
Again, I'm interested in the local, specific cases, rather than in general contract theory :-). I fully agree that some peers may want to have contracts between each other, and the associated bilateral peering, and I've no problem with that whatsoever. That doesn't mean, however, that route servers are immediately removed from the discussion, it just means ISP's have different policies for routes learnt from the route server vs from bilateral peers. It is, as always, a value thing - you're getting a bunch of routes with little effort from the route servers - how much value do you get from that, vs the effort involved with initiating bilateral contractural arrangements with specific providers. How much do you trust the routes from each source? How important do you rank the need for legal comeback if stuff goes wonky? Each provider gets to make their own judgements on these things.
I ordered the ASNs for the APE and WIX route servers, back in the day.
Not strictly - I obtained the WIX ASN by mine own fair hand, it was the APE one you obtained (and for which we are eternally grateful, since it doesn't cost us anything :-).
I wouldn't have necessarily bothered if I thought they were a bad idea
Nor I :-). Cheers Si - To unsubscribe from nznog, send email to majordomo(a)list.waikato.ac.nz where the body of your message reads: unsubscribe nznog