Study On Dynamic Routing And Other Key Techniques Of Optical Transport Network For Cloud Services

With the rapid development of cloud computing technology and the large-scale construction of data centers worldwide, the traditional Internet service model is being replaced by the cloud service model gradually. Cloud services will become one of the mainstream services carried by the future optical transport and access network. Cloud services and traditional Internet services have obviously different characteristics in service types, traffic statistic features and the diverse requirements of network quality of service (QoS). The optical transport and access network is facing huge challenges in effectively carrying the cloud services. The challenges include the optimization of network resource utilization and the realization of fast and reliable end-to-end transmission. The key techniques of optical transport network for carrying cloud services have become one of the hot research issues in both academia and industry.This dissertation studies on the key techniques, including network control architecture, routing, signaling, and key algorithms, of the optical transport network for carrying the cloud services among data centers. It also studies on the QoS strategy of optical access network for the cloud services of residents. A series of innovative research results have been achieved. The main innovative contribution and work of this dissertation are listed as follows:(1) Cloud services among data centers present obvious different characteristics in traffic statistic features and diverse requirements of network QoS. They put forward higher requirements on dynamic and flexibility of optical transport network. It is a key technique problem for optical transport network to achieve elastic resource provision, flexible function extension and dynamic network reconfiguration. To solve the above problems, this dissertation presents an optical transport network control architecture with Stateful Path Computing Element (S-PCE) Cloud and Generalized Multiprotocol Label Switching (GMPLS). Cloud computing technologies (e.g., virtualization, parallel computing and shard storage) are introduced to S-PCE cloud for providing elastic resources provision and function extension. Meanwhile, we provide the multi-domain routing scheme for dynamic routing provision of cloud services with different characteristics and QoS requirements. Simulation results show that compared to the distributed architecture, the proposed S-PCE cloud and GMPLS control architecture can reduce 41.7% path provision latency in average when the inter-domain requests are much more than intra-domain requests, and reduce 10% network blocking probability when the PCE failures.(2) The new real-time dynamic cloud services put forward higher requirements to the performance of dynamic routing provision for optical transport network among data centers. However, with the development of the optical transport network toward larger scale and wider coverage, complicated network environment and unstable link state make the increase of the crankback frequency and path provision latency. To solve the above problems, this dissertation not only presents serial and hop-by-hop signal schemes, but also presents a novel bilateral-recursive region re-routing crankback scheme (BRCB) based on S-PCE cloud. In this scheme, the backtracking nodes re-route and update the region routing paths which bypass the crankback and re-routing failure nodes when crankback occurs. It not only reduces the influencing factors of network scale, signaling crankback position and frequency to lightpath setup latency, but also avoids the backtracking of teardown messages and saves the control overhead. Theoretical analysis and simulation results demonstrate that at most the proposed scheme can pay only 2.5% of the cost for network consumption and bring in 48.6% improvement in path provision latency.(3) In terms of the traditional multi-point to multi-point (MP2MP) routing and wavelength assignment (RWA) problems, researchers usually assume that the optical networks are in a single domain. However, the optical transport networks among data center develop toward multiple domains and larger scale in practice. The MP2MP routing algorithms proposed in single network domain have problems including the waste of network resources, low success probability of multiple QoS constrained MP2MP routing and large computational overhead. Meanwhile, MP2MP RWA in multi-domain network introduces problems including optimal multicast domain sequence selection and the selection of domains for core nodes. To solve the above problems, a heuristic algorithm based on network virtualization and weighted clustering algorithm (NV-WCA) are proposed. Simulation results show that compared to the single domain algorithms, the proposed algorithm can reduce 7.6% wavelength occupation, bringing in 15.3% improvement for routing success probability, and reducing 41.7% path provision latency in average.(4) The optical access network for cloud services of residents not only accesses to conventional Internet services, but also to the new cloud services with harsh QoS requirements. The optical access network cannot accurately identify cloud services and perform effective differentiated QoS guarantee. To solve the above problems, first, this dissertation proposes cloud services recognition scheme based on both port and key words for cloud optical access network. Second, we study and design QoS guarantee schemes including dynamic traffic balancing strategy and combined hierarchical queue scheduling scheme for downstream of cloud optical access network. Moreover, the performance is demonstrated by experiments on 40G coherent PON. Simulation results show that the transmission delay of three different cloud services in 40km PON are 207μs,1.1ms and 2.1ms, and the jitter are 1.3μs,79μs and 121μs. The results can meet the QoS requirements of cloud services in PON.