University researchers built Gigaflow, improving cloud data center traffic

Researchers at the University of Michigan have developed Gigaflow, a new temporary memory storage method designed to improve cloud data center traffic management. It is aimed at improving performance in programmable SmartNICs used for AI and machine learning workloads. Gigaflow works by caching shared rule segments instead of whole data packets, which enhances cache efficiency. This approach leads to a 51% higher cache hit rate and 90% fewer misses, compared to previous methods. This new system also provides a 450 times larger rule space, while using 18% fewer cache entries. The findings were presented at the International Conference on Architectural Support for Programming Languages and Operating Systems.