DPDK profiling materials

[Roman_K_Intel]

DPDK is a set of libraries and drivers for fast packet processing.

Methodology

Benchmarking and Analysis of Software Data Planes (slides)

Tutorials

Profile Data Plane Development Kit Code with Intel® VTune™ Amplifier (article)

DPDK-on-the-Go – Profile DPDK Applications on Your Windows Laptop (article)

Data Plane Development Kit: Performance Optimization Guidelines (article)

"Profile Your Application" topic in DPDK Programmer’s Guide

DPDK Application Profiling using Intel® VTune™ Amplifier (video)

Microarchitecture analysis

Top-down Microarchitecture Analysis Method (Cook Book article)

Improving Driver Performance – A Worked Example (slides)

Analyzing Open vSwitch* with DPDK Bottlenecks Using Intel® VTune™ Amplifier (article)

I/O analysis

I/O Issues: Remote Socket Accesses (Cook Book article)

Core Utilization in DPDK Apps (Cook Book article)

PCIe Traffic in DPDK Apps (Cook Book article)

Virtual Machines

Profiling KVM Kernel and User Space from the Host (VTune documentation)

Other

Instrumentation and Tracing Technology APIs (VTune documentation)

Attaching ITT APIs to a Launched Application (VTune documentation)

[Roman_K_Intel]

VTune Cook Book recipe "I/O Issues: DPDK Empty Cycles" was published.

This recipe covers:

• How to trace "DPDK Empty Cycles" (cycles when DPDK does not fetch any packets) using ITT API (optionally in attach mode).
• How to use Custom Collector script to gather and visualize workload-specific data in CSV format.

Please share your feedback if this feature is useful to analyze DPDK workloads.

[Roman_K_Intel]

The new article Profiling KVM Kernel and User Space from the Host was published in Intel® VTune™ Amplifier 2019 User Guide.

It covers how to profile both Host and KVM virtual machine simultaneously. This may be important for performance analysis in NFVI (Network Function Virtualization Infrastructure).

Please share your feedback if this VTune feature is useful for your workloads.

[Roman_K_Intel]

VTune Cook Book recipe "Core Utilization in DPDK Apps" was published.

This recipe introduces:
- DPDK Rx Spin %: percent of rte_eth_rx_burst calls fetched zero packets / total number of calls (per Thread)
- DPDK Rx Batch Histogram: statistics how many packets were fetched each call (per Port/Queue/Thread)

Please share your feedback if this VTune feature is useful for your workloads.

[Roman_K_Intel]

VTune Cook Book recipe "PCIe Traffic in DPDK Apps" was published.

This recipe introduces PCIe Bandwidth metrics used in the Intel® VTune™ Amplifier to explore the PCIe traffic for a packet forwarding DPDK-based workload:

Inbound PCIe Bandwidth caused by device transactions targeting the system memory

• Inbound Reads show device reads from the memory
• Inbound Writes show device writes to the memory

Outbound PCIe Bandwidth caused by CPU transactions targeting device's MMIO space

• Outbound Reads show CPU reads from device's MMIO space
• Outbound Writes show CPU writes to the device's MMIO space