To accommodate explosive data traffic, today's data centers are considering introducing optical switch-based networking technologies that have characteristics of low latency, low power, and ease of expansion compared to conventional electrical switch-based networking technologies with a hierarchical architecture. However, circuit-based optical switching has the disadvantage of lacking flexibility in the optical network by providing a limited number of optical paths in a point-to-point manner. This switching also requires a dual management of packets and circuits due to the long reconfiguration time of circuit switching. In addition, packet-based optical switching has attempted to accommodate layer-2 packets with optical technology, but has faced practical limitations such as the complexity of optical control processing and the difficulty of buffering with forward-and-store schemes (i.e., fiber delay line and O-E-O loopback buffer) in optical switches. In this thesis, to address these problems, we present a packet-switched optical network (PSON) architecture design with a photonic frame wrapper that can generate variable-size photonic frames for optically interconnected intra-DCN. The proposed architecture applies a store-and-forward buffer scheme (i.e., virtual output queue based input buffer) and can increase network utilization by minimizing guard-time insertion in the optical switch domain.
In this thesis, we report on an experimental realization of a PSON that is capable of variable-size photonic frame transmission in intra-DCN. The experimental results of PSON are also presented to demonstrate the feasibility of the PSON solution as an integrated system. In reality, the traffic in DCN is non-uniform because a variety of applications are deployed and placed non-uniformly across racks. The conventional maximal matching scheduling algorithm guarantees 100% performance in uniform traffic, but is not effective in non-uniform traffic. To improve the performance of non-uniform traffic, we introduce a new scheduling algorithm called coordinated-round robin pointer (C-RRP). The C-RRP achieves higher throughput and lower latency for non-uniform traffic than conventional algorithms. Inter-DCN communication degrades network efficiency due to long delays. As a result, the dynamics of the tunable laser are degraded in the proposed PSON where one input tunable laser is shared by N destination outputs. To overcome this, we introduce a photonic frame wrapper with proxy solution to fully exploit the gain of the proposed architecture even in inter-DCN communication with long delays, which is compliant with standard TCP instances at end nodes. In inter-DCN communication, the proxy solution demonstrates a noticeable TCP performance enhancement compared with the photonic frame wrapper in terms of bandwidth efficiency and end-to-end delay by theoretical and OPNET simulation analyses.