Study On Key Techniques In Passive Optical Networks And Optical Grid Applications

The bandwidth-intensive applications, such as interactive video and multimediaservices, have further increased the demand of bandwidth. Thus wavelength divisionmultiplexing passive optical network (WDM-PON) is viewed as a promising candidate torealize the next generation optical access networks due to its dedicated bandwidth for eachsubscriber and more flexible bandwidth management. The first half of this thesis willcover three technologies in a WDM-PON, including multicast overlay scheme, automaticprotection switching scheme and tunable Fabry-Perot laser diode (FP-LD) self-seedingscheme. In the second half of the thesis, WDM optical network is utilized as a promisingvirtual computing environment, which connects widely distributed computing resources tosupport large-scale scientific, engineering or commercial computing applications. It is socalled “optical Grid” system. Grid applications may range from the simple transfer of alarge data set to the complex execution of a collection of interdependent tasks. Especially,we for the first time introduce the optical Grid applications into WDM optical accessnetworks to realize massive data aggregations. However, for such a system involvingmany heterogeneous computing and network resources, faults seem to be inevitable. Thisthesis will also address the issue of maximizing grid application availability in real-timeoptical Grid systems through resource-fault-tolerant scheduling techniques.We will briefly discuss our works in the following sub-topics.1. Multicast Overlay Scheme in WDM-PONsMulticast overlay scheme in WDM-PONs can be realized by establishing one-to-many light paths on the optical layer, and thus reduces the loading of the electronicnetwork processors or routers on the network layer and achieves much higher processingspeed. In this thesis, we have proposed our two feasible schemes to overlay multicasttransmission onto the existing point-to-point traffic in a WDM-PON.1）In the first approach, by using a dynamic wavelength reflector in each WDM channel,OLT selectively enable the multicast data imposed on the corresponding downstreamunicast carrier. Several different configrations of the wavelength reflector are presented, which is composed of some mature and simple devices. The reconfigurablemulticast control is centralized in the OLT and is transparent to all ONUs. No lightsource is needed in the ONUs by employing re-modulation technique.2）The second approach uses a dual-parallel MZM in each wavelength channel togenerate the optical sidebands for multicast DPSK data modulation. The downstreamNRZ unicast data is carried on the optical baseband carrier, which will be reused forthe upstream NRZ data at the respective ONU. By simply switching the RF controlsignal on or off, the centralized multicast function can be reconfigured dynamically. Asthe upstream signal and the multicast signal are carried on different subcarriers, thoughon the same fiber path, the Rayleigh backscattering effect is much alleviated.2. Protection Switching Schemes in WDM-PONsA survivable WDM-PON architecture which provides automatic protection switching(APS) capability can avoid enormous loss in data and business due to fiber cuts. We willpropose and demonstrate two novel survivable WDM-PON architectures:1）Centrally-controlled intelligent protection scheme in a single WDM-PON: Bymonitoring the optical power of each channel on both the working and protection paths,the proposed scheme can perform an effective protection switching with the aid of theproposed logic decision unit in more practical operation scenarios. The scheme candeal with both the feeder fiber and the distribution fiber failures.2）Cross-protection dual-PON-based architecture with colorless ONUs: It can provide1+1protection for downstream traffic and1:1protection for upstream data againstboth feeder fiber and distribution fiber failures by using the fiber links and AWGs ofthe neighboring WDM-PON. It has the minimum number of extra protection fibers,much improved wavelength utilization and better transmission performance comparedwith the other existing protection schemes.3. Upstream Multi-Wavelength Shared PONProviding cost-effective, smooth capacity upgrades while maintaining compatibilitywith existing PON standards will be of great concern for network operators. A novelUMWS-PON will be presented and demonstrated based on the proposed three possibleconfigurations of the tunable FP-LD self-seeding module at ONU. The PON not only upgrades easily upstream bandwidth capacity by introducing multiple wavelengths, butalso improves significantly bandwidth utilization by sharing all wavelength resources. Wealso study the impact of channel SL on the network performance. We for the first timeintroduce the optical Grid applications into optical access networks, more especially, intothe proposed UMWS-PON, to realize large data file aggregations. The Time-WavelengthCo-Allocation (TWCA) Problem is defined and an effective greedy approach to thisproblem is presented for the aggreagation applications. The simulations demonstrate thatthe performance is improved significantly by using our proposed “Longest File First-First-Fit first (LFF-FF)” algorithm.4. DAG Applications in Optical GridsBecause optical grid systems generally involve many heterogeneous computing andnetwork resources, network or Grid resource faults seem to be inevitable. We haveproposed two resource-fault-tolerant scheduling schemes for optical Grid applications:1）The first scheme focus only on optical link failures, the proposed Availability-DrivenScheduling (ADS) scheme improves the availability iteratively under the applicationdeadline requirements. Its performance advantages will be stated clearly in thecomparison with two other DAG scheduling schemes in different network scenarios.2）The second one addresses the case of grid resource failures in optical Grids by using aprimary-backup approach. It allocates simultaneously two copies of each computationtask to two different Grid resources. Through simulation results, we can see that itimproves greatly application availability and induces less the overhead in schedulinglength when the more network resources are available.In the end, a brief summary of all discussed topics in this thesis is given. The maincontributions of this thesis and several further studies or worth studies are pointed out atthis part.