Multi-dimensional Optimization For Resource Allocation In Optical Datacenter Networks

With the exponential increasing of Internet traffic,new-type applications and requests,e.g.,video conferencing,virtual reality(VR),holographic con-versation,big data,cloud computing,etc.,are booming.These applications with high-bandwidth,low-latency,high-reliability requirements make current carrier network and datacenter facing serious challenges.On the one hand,network operators are trying to upgrade networks and expand the capacity of datacenters to support more traffic,which leads to high capital expenditure.On the other hand,it is desirable to develop reasonable resource allocation and optimization strategies,which can improve network performance in terms of resource efficiency,throughput,etc.,for both current or newly-built networks.Besides,as natural or man-made disasters happen frequently,how to improve survivability of networks is also an important issue for network operators.It becomes more essential for optical networks,where any fiber link failure may cause huge loss of data and revenue.Hence,it attracts wide attention of carriers and service providers to solve the problem of how to improve resource efficiency and survivability in the optical datacenter network.This dissertation focuses on the resource optimization strategy with multi-dimensional coordination in optical datacenter networks.With the objective of optimizing resource usage for different kinds of requests specifically,the strategy tires to improve the capacity of the optical datacenter network to support various traffic and resource efficiency(e.g.,bandwidth and storage).Also,it can guarantee specific requirements of requests,e.g.,latency and survivability,can be satisfied,and meantime,network performance can be improved.This dissertation includes the following aspects:(1)A resource optimization strategy for both spectrum resource and re-quest type is investigated and proposed.In the optical datacenter networks,many conventional end-to-end(one-dimensional)and multicast(two-dimensional)re-quests exist at the same time,which require specific optimization methods.The strategy can serve different kinds of demands with different spectrum allocation methods,in which case,the factor that content distribution among datacenters is also considered.Distributed sub-light-tree are constructed to serve multicast requests.Furthermore,to improve the survivability,we investigate multiple protection methods,with segment protection,path protection,and shared pro-tection employed to improve spectrum efficiency.(2)An optimization strategy for both spectrum and storage resource is investigated and proposed.With the content distribution among multiple data-centers and requirements of demands considered,the strategy can adjust content dynamically according to network states,e.g.,distribution and type of different demands and network latency,to improve network survivability and decrease spectrum consumption and blocking probability.Also,it aims to reduce the network latency and cost caused by storing the same content among multiple datacenters.Hence,the trade-off of spectrum and storage resource consumption can be achieved.In this strategy,multi-path protection and content connectivity are combined to improve both spectrum efficiency and network survivability.(3)An optimization strategy for both spectrum and computing resource is investigated and proposed.Network function virtualization(NFV)is emerging as a promising technology to be employed in many areas,e.g.,cloud computing,5G,etc.NFV can improve the flexibility of resource allocation,but makes the problem of resource optimization more complicated.This dissertation proposed an efficient virtual network function(VNF)placement and scheduling scheme,where attributes and requirements of service demands,e.g.,location,latency,and data size,are considered.Since a VNF can be installed in a virtual machine(VM)as a specific application,it can be placed dynamically according to real requirements,avoiding extra bandwidth consumption and latency increase caused by constrained route computing.In this strategy,with the cost of deploying VNFs and different requirements of service function chains taken into consideration,both bandwidth and computing resource can be utilized optimally,and network performance in terms of latency and survivability can also be improved.