PULSE: Optical Circuit Switched Data Center Architecture Operating at Nanosecond Timescales
PULSE: Optical Circuit Switched Data Center Architecture Operating at Nanosecond Timescales
复制标题
DOI：
10.1109/jlt.2020.2997664
复制
发表时间：
2020-02
影响因子：
4.7
通讯作者：
Joshua L. Benjamin;T. Gerard;D. Lavery;P. Bayvel;G. Zervas
中科院分区：
文献类型：
--
作者：
Joshua L. Benjamin;T. Gerard;D. Lavery;P. Bayvel;G. Zervas
我们介绍了脉冲，一个分$\mu$s的光路交换数据中心网络架构控制的分布式硬件路由器。PULSE是一种扁平架构，使用基于并行无源耦合器的广播和选择网络。我们采用了一种新的收发器架构，用于动态波长时隙选择，以实现重新配置时间下降到O（100 ps），建立O（10 ns）的时隙。一种新的调度算法，具有2.3 ns的时钟周期执行多次迭代，以最大限度地提高吞吐量，波长使用和减少延迟，提高整体性能。为了扩展，单跳PULSE架构使用子网络，这些子网络通过为64个节点机架中的每个节点使用多个收发器而不相交。在重新配置电路期间（epoch = 120 ns），调度算法被示出为实现高达93%的吞吐量和64个波长的100%的波长使用率，引起范围从0.7-1.2 $\mu$s的平均延迟，最好情况下中值为0.4 $\mu $s，尾延迟为5 $\mu$s，受限于时隙（20 ns）和时期大小（120 ns）。我们展示了4096节点PULSE架构如何允许高达260 k的光通道在子网中重复使用，实现25.6 Pbps的容量，使用相干接收器时能耗为82 pJ/bit。
We introduce PULSE, a sub-$\mu$s optical circuit switched data centre network architecture controlled by distributed hardware schedulers. PULSE is a flat architecture that uses parallel passive coupler-based broadcast and select networks. We employ a novel transceiver architecture, for dynamic wavelength-timeslot selection, to achieve a reconfiguration time down to O(100 ps), establishing timeslots of O(10 ns). A novel scheduling algorithm that has a clock period of 2.3 ns performs multiple iterations to maximize throughput, wavelength usage and reduce latency, enhancing the overall performance. In order to scale, the single-hop PULSE architecture uses sub-networks that are disjoint by using multiple transceivers for each node in 64 node racks. At the reconfiguration circuit duration (epoch = 120 ns), the scheduling algorithm is shown to achieve up to 93% throughput and 100% wavelength usage of 64 wavelengths, incurring an average latency that ranges from 0.7–1.2 $\mu$s with best-case 0.4 $\mu$s median and 5 $\mu$s tail latency, limited by the timeslot (20 ns) and epoch size (120 ns). We show how the 4096-node PULSE architecture allows up to 260 k optical channels to be re-used across sub-networks achieving a capacity of 25.6 Pbps with an energy consumption of 82 pJ/bit when using coherent receiver.