| Title | : | Network Control Plane Architecture and Protocols |
| Speaker | : | Anix Anbiah (IITM) |
| Details | : | Tue, 28 Aug, 2018 3:00 PM @ ALC Conference Room |
| Abstract: | : | The control plane of a data network is analogous to its brain, while
the data plane accomplishes the job of forwarding data. As in most
other areas of computing, the architecture of the control plane can be
centralized or distributed. A data network is built using network
elements (switches, routers etc.) that typically have computing
capabilities, making the physical architecture of the network
eminently suitable for distribution. In spite of this, with a goal to
make networks programmable and software-defined, there has been a
shift to a centralized architecture. However, using control elements
that can be deployed on the network elements and use standard
interfaces to the data plane, it is possible to distribute the control
plane without compromising on this goal. Given this, an important
question is whether a distributed architecture is better than a
centralized architecture. The control plane comprises various
functions (control functions) each with a different set of
characteristics and is therefore not monolithic. In theory, different
architectures can be used for different control functions, depending
on the characteristic of each function.
In this talk, we will consider control functions that are better implemented using a distributed architecture and the specific characteristics that make them so. For example, control functions that need to solve intractable problems for which elegant distributed solutions exist with reasonable approximation are candidates for a distributed implementation. Invariably, distributed control functions require protocols between the distributed control elements. We will review an example of such a protocol. Moreover, we will generalize the problem of identifying the suitable architecture for different control functions by building a canonical model of the control plane and using this model to study the performance of control functions with different characteristics. |
