Dataplane Performance Optimization with Last-Level Cache Management

This demo showcases the advancements made through OPNFV VSPERF project for NFV data plane performance optimization. This optimization is achieved by managing CPU Last Level Cache by an open source resource management daemon software that provides plug-n-play policy based dynamic real time management of cache quality of service. By having the intelligence of any VNFs real-time LLC usage and corresponding it to a pre-set policy, the demo showcases automated noisy neighbor mitigation mechanisms in an environment with high degree of variability and unpredictability in throughput and latency. Such a mechanism of controlling the LLC allocation, preserves the performance of the workload of interest. The demo further addresses avoidance of sub-optimal performance due to both underutilization and over-subscription of CPUs shared LLC and introduces secure cache association mechanisms that provide cache isolation via open interfaces enabling higher degree of automated control, test and management to meet performance SLOs in a production multi-tenant NFV environment.