In vCloud NFV Edition, resource contention is evaluated based on continuous monitoring. vRealize Operations Manager uses dynamic thresholds to monitor and analyze the behavior of VNFs. Statistical confidence bands are created for each metric and its time-series, indicating an upper and lower boundary for it. The boundaries provide visibility on resource consumption of various VNFs operating in the NFVI environment. The VNF intelligence is utilized by DRS to predict future contention and capacity exhaust and rebalance the workloads in a timely manner.
In addition to the dynamic utilization and forecasting intelligence data, policies can be defined for different classes of workloads, for example:
-
Host affinity/anti-affinity rules, ensuring if tagged VMs are placed together or not on the same host.
-
Latency Sensitivity is set to high for data plane intensive workloads, pinning CPU.
-
NUMA affinity to ensure that the VM is assigned to a dedicate node.
-
Current and target host CPU profile alignment to avoid mismatch (Enhanced vMotion Capability).
Distributed DRS can be used to balance VNF components based on their performance needs and capacity availability, eliminating resource contention that might occur. Forecasts are generated every five minutes, which allows DRS to predict and act on any possible imbalances. The optimization can be enabled in a supervised or non-supervised control.
