Quote:
Originally Posted by Sephiroth
maybe cheapskate peering arrangements that buckle on peak times (like new IOS for iPhone like it was 7/8 Nov).
|
Well, the "recent" issues have all occurred over
public peering, and the arrangements there are really "let computers talk to each other and agree routing automatically among themselves using BGP". Each company pays for their own connections to the system, and the system provides automatic connectivity to everyone else at the same speeds. The system doesn't distinguish between providers or destinations and there are no specific arrangement (other than the basic allow/ignore on a provider's ACL)
In this case VM are advertising about a dozen ways into their network (and indeed have around that many physical connections at LINX) - and the public peering exchange on LINX is really just that - a giant public switch where anyone can send data anywhere it pleases. The problem here is too many other providers are sending data back into VM's network through one node and not making sufficient use of the other 8.
Assuming this wasn't caused by a temporary fault elsewhere, this is not a routing or peering issue, it's a load balancing issue. And to be fair, load balancing across two distinct networks, a half dozen locations, a dozen ports, a hundred providers and thousands of routes is a complete bitch to get right.
---------- Post added at 10:38 ---------- Previous post was at 10:37 ----------
Quote:
Originally Posted by Chrysalis
|
Haha, they linked to Craigy's ping-o-tron