More topics should be updated, but currently I do not have time...

More topics should be updated, but currently I do not have time…

Non-intrusive Network Measurement

Instead of directly collecting packet- and flow-level features, we explore the low-level hardware features ubiquitously available in modern commodity hardware, such as the CPU cores, multi-level caches, memory subsystem, and PCIe buses, which have minimal impact on normal network operations.

Performance Diagnosis

We develop predictive models to infer performance impairments and deduce the associated bottlenecks in high-speed NFV data plane.

Performance Optimization

We implement a Deep Reinforcement Learning (DRL)-based framework to pinpoint resource contentions at runtime and automatically adjust the Last-level Cache allocation to optimize the end-to-end service performance, e.g., throughput, latency, and energy efficiency.

Collaborators: Nokia Bell Labs, Tsinghua University