Refactoring Transport for the Next Century

The idea of “refactoring ” transport protocol functionality and implementing it in a more flexible and extensible way has been the subject of a good deal of academic research, going back at least 20 years [4, 1, 3, 8, 13, 10].1 Unfortunately, none of the proposed solutions has achieved significant deployment in the global Internet. (Although various frameworks for building web services— SOAP, TWISTED, etc.—are in pretty widespread use, they are generally language-specific and run “above ” the application-layer protocol.) One plausible explanation is that the perceived value of the proposed framework(s) has never been sufficient to offset the cost of deployment. That may be changing, due to several factors including increasing diversity of application needs, more challeng-ing operational environments, and evolving security considerations (e.g., pervasive monitoring). As noted by the workshop organizers, a second barrier to evolving the stack is the presence of components in the infrastructure (middleboxes) that actively enforce assumptions about protocols above the Internet layer—effectively precluding the use of transport protocols other than TCP and UDP (and possibly SCTP). The prevalence of such elements in the network is now simply a fact of life, and may actually be increasing. However, this problem seems to have received less attention from the academic research community. The workshop organizers have it exactly right: the prob