Towards Content-Centric Geometric Routing
Wouter Taverier*, Sahel Sahhaf*, Didier Colle*, Mario Pickavet* and Piet Demeester*
* Ghent University, IBCN-iMinds, INTEC, Gaston Crommenlaan 8 bus 201, B-9050 Gent, Belgium
Email: {wouter.tavernier.sahel.sahhaf.didier .colle.mario.pickavet.piet.demeester}@intec.ugent.be
Abstract-Content delivery is a crucial feature of existing cloud
and telecom networks. This is confrmed by the tremendous
success of media streaming services such as Spotify and Netfix,
as well as the content and fle-distribution systems such as
BitTorrent. A recurring problem in these type of network services
is about keeping the protocol overhead as low as possible while
maximizing the efciency of such systems in terms of network
delay to customers.
In this paper we propose the use of a routing system-inferred
coordinate system to improve: i) content server selection upon
receiving content requests, and ii) the mapping of content to
servers/caches. We describe the required protocol mechanisms,
and evaluate potential gains using coordinates of Geometric
Tree Routing and compare it to pure IP-based mechanisms or
measurement-based content systems relying on coordinates. The
proposed approach can be further extended in order to include
alternate geometric systems for example supporting hyperbolic
geometries.
I. INTRODUCTION
Content delivery has become an essential aspect of current
cloud and telecom networks. The wide range of media services
such as Spotify, Netfix, or content distribution systems such as
BitTorrent are omnipresent on our smartphones, tablets, smart
tvs and pcs. Most of these services somehow rely on caching
and replication. However, being proprietary and deployed
in isolation, these content distribution technologies do not
allow to uniquely and securely identify named information
independently of the distribution channel. In many cases,
these technologies are implemented as an overlay, leading to
needless inefciency.
Content-Centric Networking (CCN) [1] aims at enabling
data to become independent fom their network/physical lo
cation, application, storage support but also their transport en
abling in turn in-network transparent caching and replication.
CCN design focusing on name-based data access therefore
gets increase interest from the network research community.
The conventional CCN concept of proposing name-based
data access at the network level where content names get
overloaded with a locator semantic exacerbates the memory
scaling limits of shortest path routing already observed in
many experimental studies and theoretic investigations. This
approach is confonted with the issue that these name spaces
have not been designed to sustain forwarding performance and
scaling compared to IF address space which can be effciently
aggregated, summarized and translated.
This paper argues that efcient and scalable content-based
forwarding requires to keep addresses or locators (instead of
names) as information unit for routing/forwarding decision.
978-1-4799-8030-7/14$31.00 ©2014 IEEE
The name of a resource indicates what we seek and an address
(locator) indicates where it is. In addition, the selection of the
addressing space for CCN should also elevate the memory
scaling problem induced by shortest path routing. Therefore,
we propose an alternative model for Content-Centric Net
working (CCN) in which content names are assigned locators
and content routing is performed on these locators. Routing
decisions being performed on locators, this model avoids
name-to-locator resolution by intermediate nodes but instead
considers that name resolution provides suffcient information
for the receiver to determine the best route towards the selected
destination.
The rest of this paper is structured as follows. Section II
details the architecture of the proposed CCN approach and
provides elements of the design and diferent steps of the
content-centric geometric routing approach. Next, the concept
is evaluated in a simulation setup, as well as validated at a
lower level through emulation, both documented in Section
III. Section IV recapitulates the learned lessons and formulates
some potential future work.
II. ARCHITECTURE AND DESIGN
A Content-Centric Networking system consist of two fun
damental functions: the localization function and the routing
function. The localization function performs by using infor
mation name as input to determine the corresponding locator,
say x. The ruting function (distributed by nature) determines
locally and independently for any destination the adjacent
node along a loop-free path such that incoming messages
directed to this destination can reach it. Obviously, by unifying
the interpretation of the locator x and its interpretation of
such information as distance d fom the end-point or terminal
perspective, a distance metric can be defned. Putting all these
ingredients together, leads to consider a locator space that can
be processed at the end-points by a localization function and
at intermediate nodes by a routing function.
A. Geometric ruting and coordinate systems
It remains to determine which locator value space would
best ft this role. An obvious choice would be a so-called
topology dependent label space. However, such value space
is prone to renumbering even in case of non-local topological
change; hence, unsuitable. Ideally, geometric space whose
values are coordinates assigned in a way which closely refects
the topology-inferred graph distances (e.g., hop count or
weighted path length) while being robust to topology changes.
Several attempts have been made in order to identify geo
metric spaces for applying greedy ruting in order to reduce