Summary

Today’s network subsystems are a major performance bottleneck on the paths that interconnect multiprocessor servers. 

In comparison with processing speeds, parallelization, and storage speed-ups, the capacity of network links has relatively worsened over time, and to a dramatic extent. This bottleneck has remained unaddressed due to commercial incentives focused on incremental technology advances across multiple, independent market siloes in network and server technology. This has made network interface cards (NICs), which bridge the network/server boundary, an afterthought in both technology marketplaces.

The Fast Network Interface Cards (FastNICs) program will focus on overcoming these mismatches in computing and network subsystem performance through the development, implementation, integration, and validation of novel, clean-slate network subsystems. 

The resulting experimental systems aim to speed up applications such as the distributed training of machine learning classifiers by 100x.

More information is available in the FastNICs BAA.