Why the "End to End Arguments in System Design" is wrong.

(This thought just hit me so I wanted to capture it immediately.) Briefly, with more to come later, a federated architecture, e.g., an internetwork, requires not only the federation of heterogeneous and autonomous endpoints (i.e., hosts), it requires the federation of heterogeneous and autonomous "subnets." Now the Internet enabled heterogeneous link architectures, which it unified with IP transport. But according to the end to end arguments all other "unification" should done at the end points, e.g., TCP for reliable sessions, SSL for secure sessions.

But the end to end arguments directly contradict the Internet's design goal of being a "network of networks"! The Internet is not fully a "network of networks" because it does not enable network services such as security, reliability, transactionality, etc. to be federated across heterogeneous instances of such services into a unified end to end service. The end to end arguments claimed that such services were not "network services" but "application services". This merely begs the question of how to build a federated "application level" network. Web services is the answer to this question.

I'm surprised I haven't run across this argument against the end to end arguments given how compelling it is. Has anyone run across it?

I suppose this counts as the follow up I promised in an earlier entry on the End to End.