Efficiency Optimization for Software Defined Networking

PhD Thesis Proposal Defence


Title: "Efficiency Optimization for Software Defined Networking"

by

Mr. Zhiyang SU


Abstract:

The rapid growth of cloud computing, network virtualization and big data 
brings new challenges for computer networks. Traditional network 
architectures are ill-suited to meet the requirements of enterprises and 
users. By decoupling the control plane and the data plane, Software 
Defined Networking (SDN) is proposed to enable network innovation with 
unprecedented programmability. Specifically, the flexibility of SDN 
significantly facilitates network measurement and monitoring.

Despite these advantages, many issues remain unresolved in SDN. Firstly, 
as a forwarding component, the performance of the logically centralized 
controller is a major concern. Since the number of mice flows contributes 
to a considerable fraction of the total number of flows, the frequent 
invocations of the controller may result in a high flow setup latency. 
Secondly, to perform software defined measurement tasks, the precious TCAM 
entries are required to monitor corresponding flow set. Moreover, the 
controller collects flow statistics from the switches by periodic polling, 
which incurs measurement overhead on the network. The challenge is how to 
conduct high-accuracy software defined measurement at minimum cost.

To address the first challenge, we propose CheetahFlow, which predicts 
frequent communication pair and proactively installs forwarding wildcard 
rules to eliminate the extra flow setup latency. Particularly, the 
blocking island paradigm is employed to reroute elephant flows to a 
non-congested path to further reduce the flow setup latency.

To resolve the second challenge, we propose several novel schemes in a 
top-down manner. In the first place, we propose COSTA, a cross-layer 
optimization for sketch-based measurement, which works in the application 
and the management layers. By sacrificing a small amount of accuracy, 
COSTA dramatically shrinks the measurement resource usage. We formulate 
the measurement task assignment problem as a mixed integer nonlinear 
programming problem, and develop a two-stage heuristic to efficiently 
assign concurrent measurement tasks. In the second place, we propose 
CeMon, a cost-effective flow statistics polling scheme which is a 
shim-layer between the management and the physical layers. We point out 
that flow statistics polling is a fundamental primitive for software 
defined measurement. Then, we propose a generic optimization which is 
compatible with all existing measurement frameworks. CeMon presents two 
monitoring schemes to achieve different levels of measurement granularity. 
In particular, the maximum coverage polling scheme collects the statistics 
of all active flows, while the adaptive fine-grained polling scheme 
targets at polling the statistics of a small set of flows. Both 
theoretical analysis and simulation results demonstrate the effectiveness 
of our proposals.


Date:			Friday, 28 August 2015

Time:                   2:00pm - 4:00pm

Venue:                  Room 2132B
                         lift 19

Committee Members:	Prof. Mounir Hamdi (Supervisor)
  			Dr. Kai Chen (Chairperson)
 			Prof. Gary Chan
  			Dr. Qiong Luo


**** ALL are Welcome ****