skills.rest
NVIDIA
NVIDIAOFFICIAL
Explore all skills by this author →
WebsiteGitHub

doca-flow-dpa-perf

Measures rule update and disable rates on the DPA-offloaded DOCA Flow path.

3.2k|370|Updated Feb 25, 2026
Search Tags:#benchmarking#doca#doca-flow#bluefield#dpa#performance measurement#connectx
One-click install
npx skills add https://github.com/NVIDIA/skills --skill doca-flow-dpa-perf
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: doca-flow-dpa-perf
Source: https://github.com/NVIDIA/skills/tree/main/skills/doca-flow-dpa-perf
Command: npx skills add https://github.com/NVIDIA/skills --skill doca-flow-dpa-perf

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Measuring the performance of the DPA-offloaded DOCA Flow rule update path requires correct hardware preconditions, device selection, and workload configuration; this Skill guides agents through invoking doca_flow_dpa_perf so the resulting Kops/sec numbers are defensible and comparable.

Core Features & Use Cases

  • Precondition and device validation: Confirms DPA-capable hardware (ConnectX-7 minimum, ConnectX-8 recommended, BlueField-3), VNF Flow mode, and PF/VF device targets before any run.
  • Workload-shape configuration: Guides selection of burst size, queue size, completion threshold, workers, hash pipe algorithm, and PSL table parameters using the shipped README and installed --help as sources of truth.
  • Smoke-before-bulk measurement and layered debugging: Enforces a small smoke run before sweeps, interprets iteration statistics (median, max, standard deviation), and diagnoses zero-ops or failed self-test runs through an eight-layer error taxonomy.
  • Use Case: A performance engineer asks for a baseline rule-update rate on ConnectX-8; the agent confirms DPA preconditions, runs a warmup-enabled smoke test, captures the four-tuple (command line, DOCA version, device, environment), and reports a reproducible Kops/sec number.

Quick Start

Use the doca-flow-dpa-perf skill to measure the DPA-offloaded Flow rule update rate on my ConnectX-8 device with a smoke run first.

Frequently Asked Questions about doca-flow-dpa-perf

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I measure DPA-offloaded DOCA Flow rule update rates?▼

Invoke doca_flow_dpa_perf on DPA-capable hardware after confirming VNF Flow mode and a PF or VF device target. Start with a smoke run using small operation and iteration counts with warmup enabled, then scale up only after iteration statistics show low variance.

doca_flow_dpa_perf vs doca_flow_perf: which tool should I use?▼

doca_flow_dpa_perf measures the DPA-offloaded Flow rule update and disable path on DPA-capable devices, while doca_flow_perf measures the host or DPU-CPU Flow path. The two numbers are not interchangeable, so always name which tool produced each result.

Is BlueField-2 supported for DPA Flow performance measurement?▼

No. The DPA-Provider library does not run on BlueField-2 or pre-ConnectX-7 hardware. The supported set is ConnectX-7 as minimum, ConnectX-8 recommended, and BlueField-3; on unsupported hardware, use the host-side doca-flow-perf tool instead.

Why does doca_flow_dpa_perf report zero ops or fail the self-test?▼

Zero ops usually indicates a device-binding or DPA-precondition failure, such as unsupported hardware, missing VNF mode, or an SF device target. Self-test failures typically mean traffic was not sent during the pause or the tcpdump filter does not match the configured ports.

Can I use subfunctions (SFs) with the DPA Flow perf tool?▼

No. Per the shipped README, the DPA devices will not run if SFs are used. Use a PF (recommended) or VF device target instead; there is no supported workaround for SF targets on the DPA path.

Related Skills

Explore More Software Engineering Skills→
dongg622dongg622

ix-gpu-tuner

Detect and correct BIOS, kernel, CPU, NUMA, PCIe, and driver settings affecting GPU performance.

Community
Intermediate
outfitter-devoutfitter-dev
28

performance-engineering

Profile and optimize software performance across languages and runtimes.

Official
Advanced
NVIDIANVIDIA
993

udf-judge-conversion

Review converted UDFs for correctness gaps and hidden fallback risks.

Official
Advanced
stolyarchukstolyarchuk

doca-expert

Guide C++ engineers in designing and optimizing DOCA Flow pipelines on NVIDIA SmartNICs.

Community
Advanced