Interactive Networks

Introduction

Interactive Networks is an environment to test and develop new ideas on visualization, analysis, network services and applications in the context of programmable networks. In this research, our aim is to find visualization and interaction methods that (1) reveal the structure of the network state, such as its robustness for example and that (2) allow optimization of the network either by human interaction or by automated adaptation programs. Our first prototype puts the human in the control loop by using an innovative graphical input device. Many aspects of the network can be controlled by the touch of a finger. Furthermore, this input device is suitable for advanced command and control applications that monitor and manage programmable networks.

The interactive networks prototype was built in collaboration with TNO ICT, UvA SCS and UvA SNE, and will be demonstrated as part of the dutch booth at Super Computing 2008 in Austin, Texas. The prototype consists of VMware virtualized infrastructures and commodity servers and are interconnected with a 1gbit network. Users can manipulate and control many parts of the network through a multi touch interface. We are currently working on architectural frameworks and novel network programming techniques to enable automatic adaptation of networks and applications.

Photo's and video material of our supercomputing demo can be viewed here.

Concepts

The User Programmable Virtualized Networks (UPVN) (see publications) architectural framework transforms the network into software, enabling years of computer science to be applied in the context of networks. This includes Distributed Transaction Processing (DTP), scientific computing, visualization and more. By tagging packets at the IP level of streams, a UPVN can exert fine-grained control over streams and individual packets, such as copying, transforming and rerouting. UPVN has lead to a specific architecture for programmable grid networks, generalized Token Based Networking (gTBN) (see publications), suitable for implementation in Grid middleware. gTBN is also the core architecture for our current programmable network infrastructure.

Multi-touch table

The "University of Amsterdam Multi-Touch Table" is an

innovative graphical input device that is operated by

nothing more than the touch of your fingers. It is large

enough to be used by several people simultaneously

which makes it well suited for collaborative and

command-and-control applications. Contact Rob

Belleman for details (see his web page).

NEtwork Technology and Touch table interaction

The UPVN architectural framework at network element level is implemented in streamline and its accompanying language FPL-3. It uses token based networking to programatically change traffic characteristics, paths and multicast trees. The network is controlled by an open peer-to-peer control application that implements network functions, such as the streamline interface, as services and is fully asynchronous. The implemented services take care of automatic service instantiation, network element discovery, monitoring and provisioning. Monitoring services implement adapters to Ntop (throughput per link), Ping (delay, jitter) and netstat (node level throughput) to name a few. The touch table service implements interaction of the touch table with any available locally implemented service or global controller service.

Left: Result of a partial discovery of our test bed on the touch table display

Right: Just slide your finger along the nodes to create a path. Multicast paths can be made by extending an existing path in a similar manner. Path creation will initiate a two-phase commit protocol while provisioning each node. On failure the full path is rolled back.

Left: Zoom into a node and see the streamline graph and components that result from path creation.

Right: Dynamically adjust the values for streamline components. In this picture, a sampler.

NEtwork Analysis and Automatic Adaptation

Extending the network and its services to the application domain leads to interesting new potential. One of the most compelling examples is the integration with Mathematica, a scientific computing environment. The figure to the right shows

a Mathematica session of the network

under control of the multi touch table. the

graph shows the discovered IP topology

and the 3d contour plot on the bottom

provides a real-time view on node

throughput.

Not only is Mathematica interesting tool for

network analysis and visualization, but it

is also a suitable tool for developing

adaptation programs to automate human

decision processes. See publications for

a paper on integrating networks with