3 ms·

While the designers of TCP/IP got an amazing amount of things right, one things they missed rather badly was the need to take the incentives and motivations of

by magila 9y ago

While the designers of TCP/IP got an amazing amount of things right, one things they missed rather badly was the need to take the incentives and motivations of the people participating in the network into account. I don't blame them for that. At the time the internet was an academic pursuit, so the idea of designing a protocol to resist breakage by self-interested commercial ISPs simply wasn't something that was on people's minds. Taking IP fragmentation as a particular example: The protocols should be designed such that inability to perform PMTU discovery prevents any connections being established at all. Designing protocols which will be robust in the face of human actors who may-or-may-not be entirely interested in doing things the intended way is incredibly difficult. To make things worse, once you make a mistake it is practically impossible to fix. Using fragmentation as an example again: Now that middle boxes which break PMTU discovery are established, it would be impractical to introduce a "fail unconditionally" protocol. People would simply blame the products which introduced such a protocol for being broken. You have to get it right from the start so that bad actors get caught out before their broken products can be deployed.

gerdesj 9y ago
"Taking IP fragmentation as a particular example: The protocols should be designed such that inability to perform PMTU discovery prevents any connections being established at all." You seem to be saying: "I'm right and you are wrong" here and I would suggest that is ill advised - TCP/IP based comms generally work despite ill advised configuration. You can bin ICMP on WAN and yet an OpenVPN connection still works, for example. "Designing protocols which will be robust" - see Internet: all of it. I suggest that the current internet protocols are a shining example of how to do engineering properly.
- paloaltokid 9y ago
  You know BGP was written on the back of a napkin, right? And no, I am not joking. I would also submit that BGP is not a shining example of well-done engineering.
  - aninhumer 9y ago
    That something was written on a napkin doesn't mean anything without context. It could mean someone made something up off the cuff, or it could mean they'd been thinking about it for months and were finally inspired during a meal and didn't want to lose the idea.
    - jimktrains2 9y ago
      Or that it started on a napkin and was refined and flushed out over time.
- magila 9y ago
  The internet still works without ICMP, until it doesn't. Worse, when it does fail you have no recourse because the network has accumulated a pile of hacks to get 99% of people's use cases working and network operators are not inclined to spend the energy to fix their networks just for you. And so you layer another hack onto the pile and move on, as does everyone else. Not only is the time spent layering hacks on hacks significant, the result is a network which is much less useful because only the most common usage patterns can be expected to work. Want to do something novel? Good luck. See also: https://tools.ietf.org/html/draft-thomson-postel-was-wrong-01 https://tools.ietf.org/html/draft-thomson-postel-was-wrong-0...

by magila 9y ago

gerdesj 9y ago

"Taking IP fragmentation as a particular example: The protocols should be designed such that inability to perform PMTU discovery prevents any connections being established at all." You seem to be saying: "I'm right and you are wrong" here and I would suggest that is ill advised - TCP/IP based comms generally work despite ill advised configuration. You can bin ICMP on WAN and yet an OpenVPN connection still works, for example. "Designing protocols which will be robust" - see Internet: all of it. I suggest that the current internet protocols are a shining example of how to do engineering properly.

paloaltokid 9y ago
You know BGP was written on the back of a napkin, right? And no, I am not joking. I would also submit that BGP is not a shining example of well-done engineering.
- aninhumer 9y ago
  That something was written on a napkin doesn't mean anything without context. It could mean someone made something up off the cuff, or it could mean they'd been thinking about it for months and were finally inspired during a meal and didn't want to lose the idea.
  - jimktrains2 9y ago
    Or that it started on a napkin and was refined and flushed out over time.
magila 9y ago
The internet still works without ICMP, until it doesn't. Worse, when it does fail you have no recourse because the network has accumulated a pile of hacks to get 99% of people's use cases working and network operators are not inclined to spend the energy to fix their networks just for you. And so you layer another hack onto the pile and move on, as does everyone else. Not only is the time spent layering hacks on hacks significant, the result is a network which is much less useful because only the most common usage patterns can be expected to work. Want to do something novel? Good luck. See also: https://tools.ietf.org/html/draft-thomson-postel-was-wrong-01 https://tools.ietf.org/html/draft-thomson-postel-was-wrong-0...