4 ms·

I don't get it... I thought the Internet was supposed to be decentralized... How is it possible that one or two companies can cause such widespread issues?

mholt 9y ago
Because it's not decentralized.
Karunamon 9y ago
It is.. but BGP is a sort of privileged system in that once you're peering with another provider, you're trusting them to not put garbage routes out. When they do, those garbage routes propagate. The decentralized-ness is a strength as well as a weakness. Much like IRC, also a decentralized system, where a rogue server (a server has privileged access) or services package (same) can cause widespread issues across the entire network.
hanbura 9y ago
Any one error only affects a portion of the network. For example a problem might take Australia or the American East Coast offline, but that doesn't significantly affect the rest of the network (unless they try to reach affected regions). That was the design goal and it works perfectly. What you want is entirely different. The European power network for example is designed for (n+1) redundancy: any one equipment failure doesn't have significant effects. If you take that a bit further you could include misconfigurations or even allow entire companies to fall out of the network. But each level of assurance requires more overprovisioning to compensate for failed equipment or lost capacity. And overprovisioning is expensive.
toast0 9y ago
BGP is such a great foot gun precisely because the internet is decentralized. There's no central repository of how to route traffic for an IP [1]. If there was, it would probably mess things up from time to time, but not to such a large extent. Instead, we just have to kind of trust BGP announcements -- especially if they come from ISPs that credibly could route anything (Level 3, other "teir 1" isps). [1] Actually there are some efforts to develop this. After all, IP allocations are essentially centralized under the five regional internet registries. There are some registries of routing information (RADB is the most well known, I think); but not all ASNs participate, and filtering routes from large transit ISPs is still a major problem.
tzakrajs 9y ago
Decentralized does not mean guaranteed availability; however, this failure shows that the internet is somewhat fault tolerant since most packets on the net were routed to the correct destination during that time.

mholt 9y ago

Because it's not decentralized.

Karunamon 9y ago

It is.. but BGP is a sort of privileged system in that once you're peering with another provider, you're trusting them to not put garbage routes out. When they do, those garbage routes propagate. The decentralized-ness is a strength as well as a weakness. Much like IRC, also a decentralized system, where a rogue server (a server has privileged access) or services package (same) can cause widespread issues across the entire network.

hanbura 9y ago

Any one error only affects a portion of the network. For example a problem might take Australia or the American East Coast offline, but that doesn't significantly affect the rest of the network (unless they try to reach affected regions). That was the design goal and it works perfectly. What you want is entirely different. The European power network for example is designed for (n+1) redundancy: any one equipment failure doesn't have significant effects. If you take that a bit further you could include misconfigurations or even allow entire companies to fall out of the network. But each level of assurance requires more overprovisioning to compensate for failed equipment or lost capacity. And overprovisioning is expensive.

toast0 9y ago

BGP is such a great foot gun precisely because the internet is decentralized. There's no central repository of how to route traffic for an IP [1]. If there was, it would probably mess things up from time to time, but not to such a large extent. Instead, we just have to kind of trust BGP announcements -- especially if they come from ISPs that credibly could route anything (Level 3, other "teir 1" isps). [1] Actually there are some efforts to develop this. After all, IP allocations are essentially centralized under the five regional internet registries. There are some registries of routing information (RADB is the most well known, I think); but not all ASNs participate, and filtering routes from large transit ISPs is still a major problem.

tzakrajs 9y ago

Decentralized does not mean guaranteed availability; however, this failure shows that the internet is somewhat fault tolerant since most packets on the net were routed to the correct destination during that time.