FaCa: Fast Aware and Competition-Avoided Balancing for Data Center Network

Nowadays, the scale of business data is expanding at an unprecedented rate. To cater to the needs of large businesses, data center networks (DCNs) have been widely deployed and are continuing to expand. However, the influx of large-scale concurrent flows into DCNs often results in network congestion due to the concurrent competition for resources. While existing load balancing mechanisms can handle concurrent competition, they often do so at the cost of time. As a result, there is currently no ideal solution that effectively addresses both time consumption and concurrent competition issues. In this paper, we present a novel load balancing solution called FaCa, which runs on the host-end in a completely software-based manner. FaCa incorporates Inband Network Telemetry (INT), leveraging traffic transmission within the network to swiftly obtain a partial global view of network load. Additionally, we propose the Flowing &Jumping algorithm to mitigate concurrent path competition by introducing an element of randomness to load balancing process. FaCa is easy to deploy and has demonstrated superior performance compared to other mechanisms. Our evaluation on production DCN reveals that FaCa incurs minimal additional time overhead while achieving better load balancing results compared to existing approaches. Specifically, it resulted in a 14.28% reduction in congestion and a 22.5% increase in host throughput.