一句话：提出可编程光子互连 Morphlux，重构服务器内加速器网络，提升多租户 ML 带宽、降低碎片并加速故障替换。
多租户 ML 数据中心常采用环面（torus）拓扑互连加速器，但固定拓扑会导致租户带宽受限、计算碎片和单点故障影响范围大。本文提出 Morphlux，一种服务器规模的可编程光子互连结构，可在服务器内动态重构加速器之间的连接。作者构建了端到端硬件原型，并在基于 torus 的先进 ML 数据中心场景中评估。结果表明，Morphlux 可将租户计算分配的带宽提升最高 66%，将计算碎片降低最高 70%，并缩小加速器故障的影响范围；在硬件测试床上，它使 ML 模型微调吞吐提升 1.72 倍，并能在 1.2 秒内用健康加速器替换故障加速器。
We develop Morphlux, a server-scale programmable photonic fabric to interconnect accelerators within servers. We show that augmenting state-of-the-art torus-based ML datacenters with Morphlux can improve the bandwidth of tenant compute allocations by up to 66%, reduce compute fragmentation by up to 70%, and minimize the blast radius of accelerator failures. We develop a novel end-to-end hardware prototype of Morphlux to demonstrate these performance benefits which translate to 1.72x improvement in finetuning throughput of ML models. By rapidly programming the server-scale fabric in our hardware testbed, Morphlux can replace a failed accelerator with a healthy one in 1.2 seconds.