Morphlux: Transforming Torus Fabrics for Efficient Multi-tenant ML

TL;DR

Morphlux is a programmable photonic fabric that enhances server-scale ML data-centers by significantly increasing bandwidth, reducing fragmentation, and improving fault tolerance, leading to faster training and better reliability.

Contribution

We introduce Morphlux, a novel hardware fabric that transforms torus networks for improved performance and fault management in multi-tenant ML servers.

Findings

Up to 66% bandwidth improvement for tenant compute

Up to 70% reduction in compute fragmentation

1.72X increase in ML training throughput

Abstract

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 data-centers 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 chip failures. We develop a novel end-to-end hardware prototype of Morphlux to demonstrate these performance benefits which translate to 1.72X improvement in training throughput of ML models. By rapidly programming the server-scale fabric in our hardware testbed, Morphlux can replace a failed accelerator chip with a healthy one in 1.2 seconds.