A preliminary implementation of a content–aware network node

A PRELIMINARY IMPLEMENTATION OF A CONTENT–AWARE NETWORK NODE

N. Vorniotakis, G. Xilouris, G. Gardikis, N. Zotos, E. Palis, A. Kourtis

!st Workshop on Multimedia-Aware Networking 2011 (WoMAN ‘11)

Contents• Introduction• Scope• Content-Awareness Enablers• Design • Experimental Testbed• Validation and experimental results• Acknowledgments - Conclusions

2 ICME 2011 Conference, WOMAN Workshop July 11 2011, Barcelona

Introduction• Multimedia content is anticipated to be increased at least by a

factor of 6 in 2012• Network nodes are currently agnostic to the content they deliver• In order for future network architectures to cope with this

environment – continue to provide fast switching and forwarding at the core – push the intelligence to the edge

• Given the constant evolution in hardware capabilities — in terms of CPU power and memory availability there is the capability to:

Provide new functionalities to the network nodes in order to make the network aware of the content being transferred hence applying specific policies or routing respectively


Scope • This work presents a preliminary design of a content-aware

network node• Discusses the main concepts and principles governing this design • Presents an preliminary implementation of an algorithm for

identification of multimedia streams over RTP protocol• Validates the proof-of-concept through experimental results


Content-Awareness Enablers

• Flow awareness– content-awareness should be performed per-flow of network

data– Use of hash tables where every active flow record, is

maintained by the network node• Mechanisms for removal of idle or zombie flows are mandatory in

order to be detected and removed, and free memory – Enables the processing of the minimum required amount of

packets– Resulting

• Smaller processing delays • Better scalability


Content-Awareness Enablers

• Traffic Classification– Current techniques exploit information taken from OSI Layer 3

to Layer 7– Many techniques combine multilayer information with

application data inspection (DPI) for accurate traffic identification

– Less invasive to privacy methods involve statistical analysis of the flow dynamics

– Methods used to classify traffic at application level include• Exact Matching• Prefix Matching • Heuristics methods• Machine learning based on statistical features


Design

7

• C-A functions were designed to be modular and scalable

• Flow Handling module comprises of • the Packet Capturer

module that captures incoming network packets

• Flow Handler that organizes incoming packets to network flows.

• Routing module comprises of: • Packet Marker and the

Routing Tables Handler

• Content Awareness Module depends on the Heuristics functions to determine the service of each flow using either Deep Packet Inspection (DPI) techniques or Port to Service Mapping for simple services identification

• Policer Module is in charge of applying the desired policies at the respective flows. This module includes also the queue schedulers that are used for traffic control (shaping, differentiation, prioritization)

ICME 2011 Conference, WOMAN Workshop July 11 2011, Barcelona

Design – Routing Module

8

• Routing Module• for every incoming flow, after the content is identified

three main decisions need to be made.• how to police the traffic at the ingress interface• how to route the flow • how to handle (shaping, conditioning,

prioritization) the flow at the egress. • This module exploits functionalities provided by the

Linux OS kernel and User Space utilities (i.e. iptables, traffic control

• Content Mapping Table that contains information on how to police and condition the flows depend- ing on content type

• A number of alternative local RIBs is used that are statically pre-assigned


Design – RTP dissector

9

• Incoming packets are traversing the processing loop, each one is checked to determine whether it belongs to an already established flow

• The actual detection algorithm is much more complex so it can be accurate on most cases since it has more passes and also takes into account RTCP data.


Design – RTP dissector

10

• Incoming packets are traversing the processing loop, each one is checked to determine whether it belongs to an already established flow

• The actual detection algorithm is much more complex so it can be accurate on most cases since it has more passes and also takes into account RTCP data.


Experimental Test-bed

11

• streaming server streams a flow of a video using RTP to the client that is considered to be a high priority service

• traffic generator is used to create a gradually increasing source of background traffic


Experimental - Cases

• Testing Scenarios– Case1 - content agnostic network– Case2 - traffic classification based

on policies and HTB– Case3 - traffic classification based

on content identification


Experimental-Results

13

Case1

Case2

Case3

• The proof of concept of the content-aware network node is proved

• In case3 the content aware features of the ingress node allow the selection of different path in the network

• The one way delay is not affected by the operation of the C-A algorithm


Acknowledgments

14

This work has been supported by the European Research Project FP7 “MediA Ecosystem Deployment Through Ubiquitous Content-Aware Network

Environments” ICT-ALICANTE Project No. 2010-2013. http://www.ict-alicante.eu


15

Thank you for your attention

Questions ?

Contact informationNikolaos Vorniotakis ([email protected])

George Xilouris ([email protected])


mailto:[email protected]

mailto:[email protected]

16ICME 2011 Conference, WOMAN Workshop July 11 2011, Barcelona

Technology

A preliminary implementation of a content–aware network node