NFV-RI (RHOSP) Certified

All known OSs nowadays feature power management policies that can dynamically scale the power (frequency) of CPU cores according to their workload's demand. This is the case for most IT and cloud workloads. However, they fail to manage in the same way user-plane network functions, which represent the largest and most notable portion of NFV workloads, because these appear to the OS as being always busy. Under the hood, user-plane VNFs/CNFs (like those based on DPDK) rely on polling to achieve very low latency, deterministic performance and zero packet drops, which keeps their cores constantly at 100 percent utilization, even when being completely idle. As a result, the OS is oblivious to the actual load such VNFs/CNFs have in reality at any time, and hence cannot do anything to throttle their power. NFV-RI, and in specific its FFL ("Frequency Feedback Loop") feature, is able a) to predict how busy user-plane functions will become in the next time window, and based on that b) to identify which is the the least CPU frequency they will need to keep processing packets without drops. Both of these tasks are solved using AI. In this way user-plane functions consume significantly less energy during off-peak periods, without their Service Level Objectives (SLOs) being compromised.