ACT: An App-Centric Transport Architecture for the Internet

In today's Internet, TCP and the sockets interface form the de facto transport architecture. This architecture, however, is now showing signs of age, accelerated by

  1. the rapid proliferation of networked applications over the past few years
  2. the rise in mobile devices, wireless technologies, and ultrafast datacenter networks.

The needs of applications are no longer well-served by TCP. (1) TCP pays no heed to an application's performance or cost objectives, (2) it does not handle roaming, intermittent connectivity, multiple interfaces, and path choice, (3) its in-order byte stream model is a poor match for application-level objects, and (4) it does not allow an application to adapt promptly to changing network conditions.

To redress these shortcomings, we are developing ACT, a new transport architecture for the Internet. The cornerstones of ACT are application-specified objectives for data streams and the use of application-level framing. ACT takes application objectives into account in determining all decisions pertaining to end-to-end data transmission (congestion control & scheduling) and mobile network (and multi-homing) selection, as well as in-network queue management strategies. ACT has two main technical focus areas:

  1. Transmission Control with Explicit Objectives: The use of Remy, a program to generate congestion control protocols. This approach takes as input models (which don't have to be precise) of the network, workload, and objectives, to automatically synthesize protocols that run online. This program serves both as a design tool to explore the best congestion control and scheduling methods to optimize objectives, and also as a way to implement real-world control protocols, if it is determined that computer-generated protocols can outperform the best human-designed protocols.
  2. Realizing app objectives in the network by extending the data plane.
  3. Mobile Transport: Determining the best wireless or mobile network using machine learning and traffic analysis to satisfy application performance and energy objectives, as well as traffic-aware techniques to reduce the energy consumption of cellular wireless protocols.

    Thanks to...

    National Science Foundation and Wireless@MIT