Events

 
MPhil Thesis Presentation
Optimizing Data Center Network for Real-Time Applications

by Mr Li CHEN

Date
 :  22 Aug 2013 (Thu)
Time
 :  2pm
Venue  :  Room 2463, 2/F (Lifts 25-26), HKUST

Examination Committee
Prof Chin-Tau LEA, ECE/HKUST (Chairman)
Prof Danny H K TSANG, ECE/HKUST (Thesis Supervisor)
Prof Li QIU, ECE/HKUST
 
Abstract
This work aims to provide predictable and guaranteed performance support for real-time applications in cloud data center network (DCN) by optimizing network topology and transport layer design.
 
Sensing enAbled Localization and Tracking (SALT) system is an integrated solution for positioning, tracking and resource allocation for CR network in TV white space, which illustrates the latency and throughput requirements of real-time applications in modern DCN. SALT localizes Secondary Users (SU) based on their measurement of Primary User signals in the TV spectrum, and also uses the SU movement information to dynamically manage the available spectrum with an optimal spectrum allocation algorithm to maximize the up-link throughput of the entire network.
 
We also present MCP, a distributed and reactive mechanism for DCN to achieve minimal per-packet delay while providing the guaranteed transmission rates to meet flow deadlines. To design MCP, we first formulate a stochastic packet delay minimization problem with constraints on deadline completion and network stability. By solving this problem via Lyapunov Optimization techniques, we derive an optimal congestion window update function which establishes the theoretical foundation for MCP. To be incrementally deployable with existing switch hardware, MCP leverages existing ECN mechanism available on commodity switch to approximate the optimal window update function.
 
For topology layer, FatTorus is proposed, which is a joint topology design of storage units and servers in cloud data center. FatTorus, by means of traffic routing rules, achieves performance isolation of flows generated by storage unit and servers, and is a first step in designing a heterogeneous server-centric network in DCN. FatTorus provides low latency interconnection for servers, as well as high-throughput networking for the storage units.
 
*** ALL ARE WELCOME !! ***