Optical accelerators are passive and active optical modules or sub-systems that provide fast and efficient data transmission for certain applications. The main challenges in implementing the end-to-end system containing optical modules are the integration with current Ethernet-based networks and the control and configuration of them. Software Defined Networking (SDN) along with cross-layer designs can overcome these challenges and provide optical accelerators seamlessly to the higher layers.
We have presented multicast traffic acceleration for intra data center networks using passive optical splitters. Our design is compatible with hybrid networks and its implementation requires only commodity optical components. Optical accelerators reduce switching energy consumption and latency comparing to transmission over electrical packet switching network.
The rapid growth in cloud-based services and video streaming is changing the traffic size and pattern in the optical transport networks. According to the studies, by 2017, the metro traffic increases by 560% and 75% of the overall traffic will be terminated within the metro network. Regional data centers are the significant contributors to this traffic. Companies employ mid-size data centers in the metro distances that actively communicate for the services such as virtual machine (VM) migration, backup, video streaming, load balancing or fault/disaster recovery. This growth and shift in the metro traffic characteristics necessitates flexible designs with fast reconfiguration and seamless communication within network layers.
We have designed a novel converged inter/intra data center architecture in metro-scale distances using commodity optical space switches (OSS). We employ OSS to route both east-west and north-south traffic by aggregating racks inside the data center and also providing uplink for inter data center connectivity. The network management in the metro region is through a unified SDN control plane. A 3-layered architecture is designed to receive and schedule service requests from the application layer, and configure optical nodes in the physical layer. We have experimentally demonstrated live migration of VMs among 3 data centers over 50 km. Our design provides seamless connectivity between the application and physical layers and faster network reconfiguration in metro regions.