A High-Radix, Low-Latency Optical Switch for Data Centers

Abstract

We demonstrate an optical switch design that can scale up to a thousand ports with high per-port bandwidth (25 Gbps+) and low switching latency (40 ns). Our design uses a broadcast and select architecture, based on a passive star coupler and fast tunable transceivers. In addition we employ time division multiplexing to achieve very low switching latency. Our demo shows the feasibility of the switch data plane using a small testbed, comprising two transmitters and a receiver, connected through a star coupler.

A High Speed Hardware Scheduler for 1000-Port Optical Packet Switches to Enable Scalable Data Centers

“…The scalability of data center optical switches is limited by the complexity of the control plane. The optical switch architecture demonstrated in [9] scales to over 1000 ports on the data plane but requires a scheduler performing dynamic wavelength assignment that is also scalable to 1000 ports and also achieves high throughput consistent with packet timescales. In this work, we presented a parallel hardware scheduler to meet these requirements.…”

“…As mentioned above, TDM is used on the data plane to reduce the overall switch latency by allowing many more transmitters to access the network in a given epoch than if the switch was routed by wavelength alone. Our previous experimental demonstration showed that up to M = 25 transmitters can share a wavelength in a given epoch [9]. In this switch prototype, there are L=100 timeslots, so each timeslot is 20 ns long, which at 100 Gb/s, corresponds to 250 bytes.…”

“…Recently, we proposed an optical switch that can scale to over 1000-ports in the data plane with the use of a passive star coupler core, tunable DS-DBR transceivers and highsensitivity coherent receivers [9]. The experiments demonstrated 25 Gb/s per port with a simplified low cost and low power receiver, but could scale to 100+ Gb/s per port by using a full coherent receiver and Digital Signal Processing (DSP).…”

“…Previous high port-count scheduling work has used a software defined networking (SDN) approach to compute routing, wavelength and timeslot allocation (RWTA) for TCP/IP flows, however, the computation takes 53ms [10] making it unsuitable for scheduling at the packet level. In this work, we propose a parallel iterative scheduler algorithm for the 1000-port optical switch [9] that can compute a schedule on packet timescales. We evaluate the scheduler scalability, demonstrating that it is viable to implement on 45nm CMOS ASIC and can compute a 1000-port schedule in 1 µs.…”

Parallel Star-coupler OCS Architectures using Distributed Hardware Schedulers

“…WDM) and synchronised timeslots (i.e. TDM) to establish communication [2]. Nevertheless, reconfiguring the switch requires a fast and scalable controller.…”