Packet Classification as a Fundamental Network Primitive

Dilip Antony Joseph

EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2009-63
May 15, 2009

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2009/EECS-2009-63.pdf

Packet classification is an ubiquitous and key building block of many critical network functions like routing, firewalling, and load balancing. However, classification is currently implemented, deployed and configured in an ad-hoc manner. Reliance on ad-hoc mechanisms make classification hard to configure, inefficient and inflexible.

In this thesis, we address the above limitations by elevating packet classification as a fundamental network primitive. We do so by introducing a new classification layer in the network protocol stack, and by defining two control plane protocols -- policy-based classifier deployment and generic classification offload. In policy-based classifier deployment, packets are explicitly redirected through the classifiers specified by network policy. Generic classification offload provides a signaling mechanism that enables different entities to collaboratively implement classification. Through prototype implementations, testbed experiments and formal analysis, we demonstrate that our solution simplifies deployment and configuration, and improves flexibility, efficiency and performance of packet classification.

Advisor: Ion Stoica


BibTeX citation:

@phdthesis{Joseph:EECS-2009-63,
    Author = {Joseph, Dilip Antony},
    Title = {Packet Classification as a Fundamental Network Primitive},
    School = {EECS Department, University of California, Berkeley},
    Year = {2009},
    Month = {May},
    URL = {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2009/EECS-2009-63.html},
    Number = {UCB/EECS-2009-63},
    Abstract = {Packet classification is an ubiquitous and key building block of many critical network functions like routing, firewalling, and load balancing.  However,        classification is currently implemented, deployed and configured in an ad-hoc manner.  Reliance on ad-hoc mechanisms make classification hard to configure, inefficient and inflexible.  

In this thesis, we address the above limitations by elevating packet classification as a fundamental network
primitive.  We do so by introducing a new classification layer in the network protocol stack, and by defining two control plane protocols -- policy-based classifier deployment and generic classification offload.  
In policy-based classifier deployment, packets are explicitly redirected through the classifiers specified by network policy. Generic classification offload provides a signaling mechanism that enables different entities to collaboratively implement classification. Through prototype implementations, testbed experiments and formal analysis, we demonstrate that our solution simplifies deployment and configuration, and improves flexibility, efficiency and performance of packet classification.}
}

EndNote citation:

%0 Thesis
%A Joseph, Dilip Antony
%T Packet Classification as a Fundamental Network Primitive
%I EECS Department, University of California, Berkeley
%D 2009
%8 May 15
%@ UCB/EECS-2009-63
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2009/EECS-2009-63.html
%F Joseph:EECS-2009-63