A couple of months ago, I was having Yet Another Argument with a Bitcoin Core contributor about the one megabyte block size limit.
I had asked: “what other successful Internet protocols impose arbitrary limits on themselves similar to the one-megabyte limit?”
His answer was “there are limits on the size of the global routing table in the Border Gateway Protocol (BGP) protocol.”
BGP is the protocol routers use to decide what to do when they get a data packet bound for an IP address like “184.108.40.206” – they send it to China (220.127.116.11 belongs to the China Mobile Communications Corporation right now).
So I went and skimmed through the BGP protocol documents . Then read the wikipedia page on BGP . And had the opportunity to talk with Justin Newton, who participated in the debates on Internet scalability fifteen years ago.
And found that there are no protocol-level limits on the size of BGP routing tables. There is no place in any BGP specification I can find that says “Routing Tables Shall Be No More Than Eleven Gigabytes Big.”
BGP is interesting, because it is a completely different way of doing decentralized consensus than Bitcoin. Instead of proof-of-work, BGP’s consensus is built on real-world trust relationships between people operating the networks that make up the Internet. That works surprisingly well most of the time, especially considering the stunning lack of security in the BGP protocol.
There are limits on routing table sizes, but they are not top-down-specified-in-a-standards-document protocol limits. They are organic limits that arise from whatever hardware is available and from the (sometimes very contentious!) interaction of the engineers keeping the Internet backbone up and running.
I haven’t been able to find a widely-used Internet-scale protocol that arbitrarily limits itself. The closest I’ve found is the Simple Mail Transport Protocol (SMTP) which has a SIZE header that a server can use to specify the maximum email message size it will accept. The arbitrary limit chosen by the SMTP designers is 99,999,999,999,999,999,999 bytes (just under 100 exabytes).
The HTTP 2.0 spec explicitly discusses denial-of-service attacks , but doesn’t impose hard limits: “An endpoint MAY treat activity that is suspicious as a connection error of type ENHANCE_YOUR_CALM” (I can’t help imagining the server as a California Surfer saying “enhance your calm, dude!)
That’s a well-designed spec. Trust that smart developers will fix scaling or denial-of-service issues as they arise– and, if for some unforeseen reason it turns out they can’t, trust that there will be either an amendment to the spec or a "best practices” document to fix the problem(s).