Automation & Ops ✓ NVIDIA · Official bluefield-armdoca-uromrdmaucxmpicontainer-deploymentremote-memory-operations

DOCA UROM Service Operations

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

FollowSkills review · FSRS-2.0
Not recommended
45/ 100 5-point scale 2.3 / 5
Trust12 / 25 · 2.4/5

The skill explicitly discloses that the service binary has no standalone authorization control and that any host able to establish Comch pairing may drive remote memory operations; it also recommends isolation. However, it lacks concrete least-privilege settings, user-confirmation gates, rollback procedures, and sensitive-data-flow handling, while deployment depends on container, NGC, and RDMA permissions, so points are deducted.

Reliability8 / 20 · 2.0/5

The documentation provides detailed configure/build/modify/run/test/debug workflows, layered error taxonomy, and failure feedback. However, this is a static review with no execution of key paths; BENCHMARK.md also states that the evaluation dataset was unavailable, so command, log, and version-pairing reproducibility cannot be confirmed. The static ceiling and uncertainty require deductions.

Adaptability10 / 15 · 3.3/5

Audience, scenarios, trigger phrases, and refusal boundaries are clearly specified, with separation between the service, host library, and RDMA skills. Chinese-language support and mainland-China network reachability are not addressed, and core use depends on NVIDIA NGC and external documentation, so environment-fit evidence is incomplete.

Convention7 / 15 · 2.3/5

Information architecture, loading order, dependency routing, limitations, and maintenance issue links are reasonably documented. However, license metadata conflicts across SKILL.md, SKILLCARD.yaml, and skill-card.md; signing, version, and scan fields remain TBD; no clear changelog is provided; and BENCHMARK.md records schema findings. Points are deducted for governance and maintenance uncertainty.

Effectiveness5 / 15 · 1.7/5

The skill clearly covers deployment, configuration, pairing, testing, and layered diagnosis, including a smoke-before-scale policy, which should reduce configuration mistakes. However, it intentionally omits runnable image tags, commands, and configuration examples, requiring consultation of live external guides; static files also lack verifiable representative outputs, so only limited effectiveness credit is justified.

Verifiability3 / 10 · 1.5/5

BENCHMARK.md, its metrics, result summary, and evals.json provide some auditable evidence. However, the evaluation dataset is unavailable, there is no committed test suite or CI coverage demonstrating the key paths, and many claims depend on external guides not included in the assessed material. Independent reproducibility is therefore weak.

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
  • The service has no standalone authorization list or token control; strictly constrain Comch-paired hosts and underlying RDMA permissions, and confirm isolation boundaries before deployment.
  • Do not treat the PASS result in BENCHMARK.md as independently reproducible test evidence; the evaluation dataset is unavailable and this review executed nothing.
  • License, source-repository, version, and signing metadata are inconsistent or TBD; governance metadata should be reconciled before publication.
  • Actual deployment depends on NVIDIA NGC and live DOCA guides; mainland-China network access, image retrieval, and documentation reachability require separate confirmation.
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 is for external operators and platform teams running the DOCA UROM Service container on BlueField Arm. The service executes remote memory operations—including puts, gets, atomics, and collectives—enqueued by a paired host through doca-urom. It covers NGC image retrieval, UCX component selection, queue sizing, DOCA Comch pairing, version alignment, logging, and RDMA-counter-based troubleshooting. It does not cover host application code, MPI/UCX integration design, or DOCA installation.

It guides the agent through service suitability checks, configuration, deployment, testing, and debugging workflows. It covers setting SERVICE_ARGS and UROM_PLUGIN_PATH, mounting the plugins/ directory, and starting or stopping the container under the BlueField OS container runtime. It also directs checks of the host-library and DPU-service version pair, Comch pairing, RDMA permissions, container state, service logs, and RDMA counters.

  1. An HPC platform team needs to deploy the BlueField-side executor for host nodes using doca-urom to offload collective communication.
  2. An operator needs to choose exposed UCX components or collective families and size the enqueue queues for a workload.
  3. A host doca-urom enqueue returns DOCA_ERROR_NOT_PERMITTED after context startup and the operator needs to isolate Comch or RDMA permission issues.
  4. Enqueues succeed but completions never arrive, requiring separation of service-side queue or handler faults from RDMA transport failures.
  5. An offloaded workload is slower than the host-CPU baseline and the team needs a smoke test before scaling or blaming the service.

What are this skill's strengths and limitations?

Pros
  • Maintains a clear boundary between the BlueField service container and the host-side doca-urom library.
  • Covers configure, build, modify, run, test, and debug workflows.
  • Treats Comch pairing and underlying RDMA permissions as the security boundary and recommends least privilege.
  • Provides layered observability through container state, service logs, and RDMA counters.
Limitations
  • Limited to BlueField Arm and assumes DOCA is already installed.
  • Ships no production-ready configuration bundles, image tags, registry paths, or templates.
  • Does not replace host-side doca-urom programming guidance, MPI/UCX integration documentation, or DOCA setup.
  • The supplied material does not include the detailed companion workflow contents, a test suite, or platform coverage evidence.

How do you install this skill?

Install the specific skill with the CLI flow documented in the repository README:

npx skills add nvidia/skills --skill doca-urom-svc --yes

It becomes available the next time the agent loads relevant skills. The exact service image, tag, and registry path are not supplied in SKILL.md.

How do you use this skill?

Use it on a BlueField Arm system where DOCA is already installed. Example trigger: "Deploy and validate DOCA UROM Service on this BlueField, checking Comch pairing, RDMA permissions, and host doca-urom version alignment."

For deployment, start with the installed skill's configure workflow in TASKS.md. Use CAPABILITIES.md for the service architecture, configuration axes, version contract, and error taxonomy. Route host-side doca-urom programming and DOCA installation elsewhere.

How does this skill compare with similar options?

This skill complements the host-side doca-urom skill: doca-urom enqueues operations on the host, while DOCA UROM Service executes them on BlueField. It runs over the doca-rdma substrate and does not replace RDMA. DOCA installation belongs to doca-setup.

FAQ

Does this skill install DOCA or the host library?
No. It assumes DOCA is installed on BlueField and excludes both DOCA installation and the host-side doca-urom API.
Does the service provide a standalone host authorization list?
No. The skill states that Comch pairing and underlying RDMA permissions define the access boundary.
What should be checked for a NOT_PERMITTED enqueue failure?
Check the Comch endpoint pairing and RDMA permission layer first; in this context the error is not a signal from a standalone service authorization layer.
Can I copy an image tag or configuration from the skill?
No. The skill deliberately omits image names, tags, registry paths, and pre-baked configurations; consult the applicable DOCA service guide and the target BlueField's capabilities.

More skills from this repository

All from NVIDIA/skills

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

Dynamo Interconnect Readiness Check

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

Dev & Engineering ✓ NVIDIA · Official

DOCA Environment Setup

Verify, prepare, and troubleshoot the DOCA environment while routing workloads to the right deployment path.

Automation & Ops ✓ NVIDIA · Official

DOCA Container Deployment

Deploy, validate, and troubleshoot DOCA service containers on NVIDIA BlueField.

Automation & Ops ✓ NVIDIA · Official

DOCA Argus Runtime Security Operations

Deploy Argus on BlueField to detect runtime threats and forward findings to an existing SIEM.

Dev & Engineering ✓ NVIDIA · Official

DOCA RDMA Initiator

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

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 PCC ZTR RTTCC Congestion Control

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

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.

Automation & Ops ✓ NVIDIA · Official

TAO Inference Microservice Deployment

Start, query, and stop TAO inference microservices across supported compute platforms.

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

DOCA Firefly Time-Synchronization Operations

Operate, configure, and troubleshoot DOCA Firefly PTP time synchronization on BlueField.

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 GPUNetIO Development Skill

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

Dev & Engineering ✓ NVIDIA · Official

DOCA DMA Development Guide

Guides hands-on DOCA DMA memory-copy development on BlueField and ConnectX systems.

Dev & Engineering ✓ NVIDIA · Official

DOCA SPCX Programmable Congestion Control

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

Automation & Ops ✓ NVIDIA · Official

DOCA BlueField Bare-Metal Deployment

Run, supervise, and troubleshoot DOCA binaries directly on BlueField hardware.

Related skills