TagNet: A Scalable Tag-Based Information-Centric Network

You are cordially invited to attend the PhD Dissertation Defense of Michele PAPALINI on Friday, October 23rd 2015 at 09h30 in room 351 (Main building)

Abstract:
The Internet has changed dramatically since the time it was created. What was originally a system to connect relatively few remote users to mainframe computers, has now become a global network of billions of diverse devices, serving a large user population, more and more characterized by wireless communication, user mobility, and large-scale, content-rich, multi-user applications that are stretching the basic end-to-end, point-to-point design of TCP/IP.

In recent years, researchers have introduced the concept of Information Centric Networking (ICN). The ambition of ICN is to redesign the Internet with a new service model more suitable to today's applications and users. The main idea of ICN is to address information rather than hosts. This means that a user could access information directly, at the network level, without having to first find out which host to contact to obtain that information.

The ICN architectures proposed so far are based on a ``pull'' communication service. This is because today's Internet carries primarily video traffic that is easy to serve through pull communication primitives. Another common design choice in ICN is to name content, typically with hierarchical names similar to file names or URLs. This choice is once again rooted in the use of URLs to access Web content. However, names offer only a limited expressiveness and may or may not aggregate well at a global scale.

In this thesis we present a new ICN architecture called TagNet. TagNet intends to offer a richer communication model and a new addressing scheme that is at the same time more expressive than hierarchical names from the viewpoint of applications, and more effective from the viewpoint of the network for the purpose of routing and forwarding.

For the service model, TagNet extends the mainstream ``pull'' ICN with an efficient ``push'' network-level primitive. Such push service is important for many applications such as social media, news feeds, and Internet of Things. Push communication could be implemented on top of a pull primitive, but all such implementations would suffer for high traffic overhead and/or poor performance.

As for the addressing scheme, TagNet defines and uses different types of addresses for different purposes. Thus TagNet allows applications to describe information by means of sets of tags. Such tag-based descriptors are true content-based addresses, in the sense that they characterize the multi-dimensional nature of information without forcing a partitioning of the information space as is done with hierarchical names. Furthermore, descriptors are completely user-defined, and therefore give more flexibility and expressive power to users and applications, and they also aggregate by subset.

By their nature, descriptors have no relation to the network topolgy and are not intended to identify content univocally. Therefore, TagNet complements descriptors with locators and identifiers. Locators are network-defined addresses that can be used to forward packets between known nodes (as in the current IP network); content identifiers are unique identifiers for particular blocks of content, and therefore can be used for authentication and caching.

In this thesis we propose a complete protocol stack for TagNet covering the routing scheme, forwarding algorithm, and congestion control at the transport level. We then evaluate the whole protocol stack showing that (1) the use of both push and pull services at the network level reduces network traffic significantly; (2) the tree-based routing scheme we propose scales well, with routing tables that can store billions of descriptors in a few gigabytes thanks to descriptor aggregation; (3) the forwarding engine with specialized matching algorithms for descriptors and locators achieves wire-speed forwarding rates; and (4) the congestion control is able to effectively and fairly allocate all the bandwidth available in the network while minimizing the download time of an object and avoiding congestion.

Dissertation Committee:

• Prof. Antonio Carzaniga, Università della Svizzera italiana, Switzerland (Research Advisor)
• Prof. Fernando Pedone, Università della Svizzera italiana, Switzerland (Internal Member)
• Prof. Robert Soulé, Università della Svizzera italiana, Switzerland (Internal Member)
• Prof. Peter Pietzuch, Imperial College London, United Kingdom (External Member)
• Prof. Christian Tschudin, University of Basel, Switzerland (External Member)