Dev & Engineering ✓ NVIDIA · Official bluefield-3roce-v2doca-pcccongestion-controldpaccdpardma

DOCA PCC ZTR RTTCC Congestion Control

Deploy, tune, and validate NVIDIA’s RoCE RTT congestion-control reference algorithm on BlueField-3.

FollowSkills review · FSRS-2.0
Not recommended
48/ 100 5-point scale 2.4 / 5
Trust11 / 25 · 2.2/5

The documentation describes scoped edits, version matching, parameter range checks, and safety-policy routing, with no evidence of credential theft or covert exfiltration. However, the skill can affect production RoCE traffic and lacks explicit user confirmation, concrete rollback steps, and complete permission/data-flow disclosure; its license claims also conflict with the NOASSERTION metadata, so points were deducted.

Reliability8 / 20 · 2.0/5

Install, build, modify, test, and debug workflows are extensive and include an error taxonomy. However, key behavior depends on omitted companion skills, installed SDKs, and specific hardware; the set-params call signature is inconsistent across files, and static review cannot verify execution, so the score is capped and reduced.

Adaptability11 / 15 · 3.7/5

Triggers, audience, scenarios, and non-fit boundaries are unusually explicit, including implicit user phrasing. The scope is nevertheless tightly bound to BlueField-3, DOCA, and live RoCE environments; Chinese support is not documented, and core public documentation depends on overseas NVIDIA services without mainland-China reachability evidence, so points were deducted.

Convention8 / 15 · 2.7/5

The loader plus CAPABILITIES.md and TASKS.md provide good progressive disclosure, examples, dependencies, limitations, routing, and maintenance intake. However, recommended Instructions/Examples sections are absent, version/signature/signing fields remain TBD, license metadata conflicts, and changelog and maintenance ownership are incomplete, so points were deducted.

Effectiveness6 / 15 · 2.0/5

The skill offers directly actionable workflows for integration, variant selection, parameter tuning, and diagnosis, with clear boundaries and routing. There is no independently reproducible representative output or execution evidence, and outcomes depend heavily on local hardware, SDK state, and other skills; static review therefore cannot establish full task completion.

Verifiability4 / 10 · 2.0/5

The files provide API descriptions, build procedures, error classifications, and a benchmark report. The benchmark dataset is unavailable, the report contains only aggregate metrics, no committed CI/test suite covers the key paths, and no third-party reproduction evidence is supplied, so only limited credit is justified.

Evidence confidence:Low Reviewed Jul 20, 2026 Reviewed revision 55f18499943e
The upstream repository has new commits since this review. The score still applies to the reviewed revision shown and may not cover the latest changes.
Before you use it
  • Obtain explicit user confirmation before execution or rebuild, and validate on non-production traffic first; no concrete executable rollback procedure is provided.
  • Reconcile the full doca_pcc_dev_set_ztr_rttcc_params signature against the installed headers and all companion documents before following the examples.
  • License, path, and version information is inconsistent or still TBD; metadata should be normalized and reproducible test evidence added before publication.
  • The skill depends on specific DOCA/DPACC/BlueField-3 environments and overseas documentation services; mainland-China package and network reachability should be verified.
Review evidence [1][2][3][4][5][6][7][8][9]
See the full review method →

What does this skill do, and when should you use it?

This skill targets developers deploying NVIDIA’s DOCA-shipped Zero-Touch RoCE RTT-based Congestion Control reference algorithm on a BlueField-3 DPU. It explains how to wire the algorithm into the shipped DOCA PCC C sample, select a DPA build-time variant, and adjust host-set parameters. It also helps diagnose initialization failures, DOCA status errors, and cases where traffic is not being modulated. It is not a standalone PCC application, a custom-algorithm tutorial, or a read-only counter-inspection skill.

Inspects the DOCA libraries, headers, and PCC application under /opt/mellanox/doca and reads the pkg-config module doca-pcc-ztr-rttcc-algo. It guides edits to the shipped PCC sample so it dispatches to doca_pcc_dev_ztr_rttcc_algo, builds the DPA-side translation unit with DPACC, selects vanilla, path-migration, RX-rate, multipath, or window-probeless variants, and changes host-side parameters through doca_pcc_dev_set_ztr_rttcc_params. It also directs validation through PCC counters and provides layered diagnosis for DOCA_PCC_DEV_STATUS_FAIL and DOCA_ERROR_* results.

  1. A developer has RoCE-v2 traffic on BlueField-3 and wants the shipped zero-touch reference algorithm as a baseline.
  2. The DOCA PCC sample already builds, and the developer needs to connect the shipped algorithm to its user-algorithm callbacks and slot.
  3. A deployment requires choosing among vanilla, path-migration, RX-rate, multipath, or window-probeless variants.
  4. The algorithm loads but counters or traffic behavior remain flat under congestion.
  5. A host operator needs to tune exposed RTTCC parameters without rebuilding the DPA-side image.
  6. A team must decide whether the reference algorithm is sufficient or a custom PCC algorithm is required.

What are this skill's strengths and limitations?

Pros
  • Has a clear scope around NVIDIA’s shipped ZTR RTTCC reference algorithm rather than generic PCC theory.
  • Covers sample integration, DPA compilation, variant selection, host-side tuning, validation, and layered troubleshooting.
  • Separates the algorithm library from the host-side doca-pcc lifecycle and the counter-inspection tool.
Limitations
  • Requires a specific DOCA, DPACC, BlueField-3, firmware, Linux, and live RoCE-v2 environment.
  • The skill is a thin loader and does not include the algorithm implementation, a standalone PCC application, or full mathematical descriptions of every variant.
  • It does not cover writing a custom algorithm from scratch, general PCC programming theory, or counter-only inspection.
  • The supplied material shows no runtime benchmark or hardware test results for this specific skill.

How do you install this skill?

Install the specific skill with the repository’s supported Skills CLI flow: npx skills add nvidia/skills --skill doca-pcc-ztr-rttcc-algo --yes. Prepare the required DOCA SDK, matched-version DPACC compiler, BlueField-3 DPA, enabled firmware custom-PCC slot, and a port carrying RoCE-v2 traffic. The supplied material does not provide detailed DOCA or DPACC installation commands; use doca-setup first when DOCA is not installed.

How do you use this skill?

After installation, trigger it with a task such as “How do I wire doca_pcc_dev_ztr_rttcc_algo into the existing DOCA PCC sample and choose the multipath variant?” or “The algorithm loads, but my RoCE traffic is not being modulated—how should I debug it?”. Read TASKS.md for install, configure, build, modify, run, test, debug, and use workflows. Read CAPABILITIES.md for the API, variants, parameters, counters, errors, observability, and safety policy. Route counter-only inspection to doca-pcc-counters.

How does this skill compare with similar options?

The skill explicitly distinguishes itself from the host-side DOCA PCC framework, the doca-pcc-counters inspection tool, and custom PCC development. Its main adoption decision is whether NVIDIA’s shipped reference algorithm is sufficient or a custom algorithm is needed.

FAQ

Can this skill run the algorithm by itself?
No. The algorithm is a DPA-side library consumed by the host-side doca-pcc framework, using the PCC C sample supplied with DOCA as its host application.
Can parameters be changed without rebuilding the DPA image?
The skill documents host-side tuning through doca_pcc_dev_set_ztr_rttcc_params, so exposed parameters can be adjusted without rebuilding the DPA-side image for each change.
What should I check if the algorithm loads but has no visible effect?
Confirm that RoCE-v2 traffic is active on the target port and inspect PCC counters to determine whether the algorithm is modulating flows. Initialization failures should be escalated through the algorithm, doca-pcc, and doca-debug layers.
Is this the right skill for designing a new congestion-control algorithm?
No. It is for deploying and evaluating the shipped reference algorithm. New algorithm development belongs with doca-pcc and the public PCC programming guide.

More skills from this repository

All from NVIDIA/skills

Dev & Engineering ✓ NVIDIA · Official

DOCA SPCX Programmable Congestion Control

Operate and evaluate SPCX congestion-control algorithms on BlueField DPA before production rollout.

Dev & Engineering ✓ NVIDIA · Official

DOCA RDMA Initiator

Guides accelerator-initiated one-sided RDMA development on the DPA datapath.

Dev & Engineering ✓ NVIDIA · Official

DOCA Flow DPA Provider

Guides developers in exporting DOCA Flow pipes and resources into BlueField DPA for direct kernel-side access and mutation.

Dev & Engineering ✓ NVIDIA · Official

DOCA PCC Custom Congestion Control

Guides host-side loading and troubleshooting of custom PCC algorithms on BlueField

Automation & Ops ✓ NVIDIA · Official

DOCA PCC Diagnostic Counters

Read fixed firmware and hardware PCC congestion diagnostics on ConnectX and BlueField devices.

Dev & Engineering ✓ NVIDIA · Official

DOCA DPA Host-Side Development

Guides BlueField host programs through DOCA DPA loading, launch, capability checks, and debugging.

Dev & Engineering ✓ NVIDIA · Official

DOCA Flow DPA Performance

Guides defensible measurement of DOCA Flow rule-update and disable rates on DPA-capable NVIDIA hardware.

Automation & Ops ✓ NVIDIA · Official

DOCA BlueField-3 Bring-Up

Guides operators through BF3 day-one bring-up, recovery, RShim/BFB provisioning, and installation validation.

Dev & Engineering ✓ NVIDIA · Official

DOCA Verbs Raw RDMA Control

Guides DOCA developers who genuinely need low-level verbs control, porting, and diagnosis.

Dev & Engineering ✓ NVIDIA · Official

DOCA STA Storage Target Acceleration

Build and debug RDMA NVMe-oF storage targets accelerated by DOCA STA on BlueField.

Automation & Ops ✓ NVIDIA · Official

DOCA Bench Benchmarking Skill

Measure DOCA library throughput, latency, and bandwidth reproducibly on real NVIDIA networking hardware.

Dev & Engineering ✓ NVIDIA · Official

DOCA GPI GPU-Initiated RDMA Skill

Helps CUDA kernels drive RDMA queues directly from GPU memory without host-CPU mediation.

Dev & Engineering ✓ NVIDIA · Official

DOCA UROM Host-Side Offload Development

Guides HPC, UCX, and MPI developers in offloading remote memory operations to a BlueField DPU.

Automation & Ops ✓ NVIDIA · Official

DOCA UROM Service Operations

Deploy and troubleshoot the DOCA UROM service container on BlueField Arm.

Dev & Engineering ✓ NVIDIA · Official

DOCA GPUNetIO WRITE Bandwidth Benchmark

Build, run, and interpret sustained RDMA WRITE bandwidth measurements driven by CUDA kernels through the DOCA GPUNetIO surface.

Dev & Engineering ✓ NVIDIA · Official

GPUNetIO RDMA Write Latency Benchmark

Guides reproducible measurement of CUDA-kernel-initiated GPUNetIO RDMA WRITE latency and tail behavior.

Automation & Ops ✓ NVIDIA · Official

Dynamo Interconnect Readiness Check

Verify RDMA, NVLink, and NIXL transport readiness before trusting performance results from disaggregated Dynamo serving.

Dev & Engineering ✓ NVIDIA · Official

DOCA RDMA Programming Guide

Helps developers build, configure, and debug DOCA RDMA applications on BlueField, ConnectX, and DOCA hosts.

Dev & Engineering ✓ NVIDIA · Official

DOCA DPA High-Level Tracer

Capture and decode BlueField DPA programming events to diagnose kernel behavior, communication latency, and trace failures.

Dev & Engineering ✓ NVIDIA · Official

DOCA GPUNetIO Development Skill

Helps developers connect CUDA kernels on NVIDIA GPUs to DOCA network queues for GPU-side packet I/O and debugging.

Related skills