Date of Award

Fall 2008

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Electrical Engineering - (Ph.D.)

Department

Electrical and Computer Engineering

First Advisor

Nirwan Ansari

Second Advisor

Cristian Borcea

Third Advisor

Kevin W. Lu

Fourth Advisor

Roberto Rojas-Cessa

Fifth Advisor

Yanchao Zhang

Abstract

After 9/11 and the 2003 power grid failure in North America, storage area network (SAN) extension has emerged as a critical option to ensure business continuity. However, SAN extension encounters challenges in the access network including the scalability, cost, bandwidth bottleneck, and throughput. In this dissertation, a new solution, SAN extension over passive optical networks (S-PONs), has been proposed to address the above problems. PONs are the mainstream wireline technology for upgrading the megabit-level access solutions (such as xDSL and Cable Modem) into gigabit-level broadband access. To tackle the scalability problem and cost challenge, the S-PON architecture has been designed based on the existing point-to-multiple-point (P2MP) PON infrastructure. To address the bandwidth bottlenecks in SAN extension, three solutions have been proposed to carrying storage signals with gigabit•level transmission. In addition, this dissertation introduces a new device, XtenOLT, which is an upgraded optical line terminal (OLT) with storage provisioning capacity. Two core functional modules, the dynamic resource management (DRM) module and the transmission module, are implemented in XtenOLT, respectively. In DRM, a new buffer management scheme, Tetris, is proposed to manage the buffer pools of XtenOLT, in order to improve SAN extension throughput and utility. Our experimental results show that, in the physical layer, the proposed S-PON transmission technologies successfully deliver SAN traffic to the long-haul at the rate of 2.5 Gb/s; in the network layer, S-PON with XtenOLT dramatically enhances deliverable throughput and utility over long-distance transmission.

The transmission module further adopts one of the mostly used transmission schemes, time division multiple access (TMDA). In TDMA S-PON, the upstream bandwidth allocation (BA) is one of the critical issues. In the past several years, numerous BA algorithms have been proposed, but most of them are presented in an ad- hoc manner, lacking a generic framework under which these algorithms can be evaluated, compared, and further improved. This dissertation proposes a novel state space model to represent the PON BA algorithms with state variables and input variables under a unified framework. Using this new model, the system level characteristics of diverse BA algorithms have been analyzed. Their performance difference in delay, throughput, and packet loss has also been analyzed from the system point of view. Within the framework of the proposed model, a suitable controller and compensator have been proposed to meet the prescribed objectives such as system robustness, accuracy, and transient performance. Lastly, the established state space model has been extended to the non-linear predictor-based dynamic bandwidth allocation scenario.

Share

COinS