Application Private Networks
What is APNet(Application Private Networks)?
Application-Private Networking is a new approach to network architecture
which chooses protocol elements, including multiplexing strategies, based
on application requirements. The approach fills the gap between programmable
networking approaches such as active networking and the requirements of
networked applications. The application's requirements are declarative, and
to the degree possible, application-centric choices of protocol elements are
made. As both network conditions and application requirements vary,
the protocol architecture itself is dynamic.
Why do we need APNet?
Two examples, the telephone network and the Internet, are useful in
understanding this architectural framework.
The telephone system's network architecture is engineered to deliver
a band-limited audio channel appropriate for interactive voice telecommunications.
The application requirements then, include, the ability to deliver
about 3000Hz of audio, with some limits on delay and audible impairments;
these requirements have been met in the telephony architecture by
using a call set-up protocol (of considerable complexity) to set up a point-to-point
channel used to carry a voice stream.
The Internet design, requiring interoperation across a variety of networks
and operating conditions, and intended to service many applications,
must choose protocols that can tolerate an extremely wide variety of network
conditions. Thus, the basic IP transport service is a minimal datagram service,
response to network dynamics such as topology changes is provided by dynamic
routing. Other, more application-specific requirements such as ordering,
reliability, etc. are provided by end-to-end overlay protocols such
as the Transmission Control Protocol, TCP.
The attempt to remove many dynamics in network conditions within the
call makes the telephony architecture limited in its ability to efficiently
handle applications with dynamics very different than that of voice.
Likewise, TCP/IP, intended to be agnostic with respect to applications
and adapt to a large (but not all-encompassing) range of network conditions,
has engineering tradeoffs, such as substantial overprovisioning (to control
delay jitter) to support applications such as voice and video.
How does APNet work?
An ideal network architecture would have the property that at any given
time, the application requirements and network conditions would result
in the best known selection and placement of protocol elements. This is shown
in Figure 1. Application requirements are enforced in both the local host
and selected nodes in the network, shown in Figure 2. The configuration of
APNet makes use of the Knowledge
Plane and interacts with KP as shown in Figure 3.
Figure 1: Dynamic Protocol Refinement
People
Jonathan M. Smith
Stefan Miltchev
Dekai Li
Yun Mao
Ben Voigt
Papers and presentations
Towards Application-Private Networks (APNets), Dekai Li,
Jonathan M. Smith, February 2003 (ps)(pdf)
Thoughts on Network Protocol Engineering, Jonathan M. Smith, November
2002 (ppt)
Links
Knowledge Plane
Last modified: April, 2003 Distributed
System Lab, University of Pennsylvania
