Networking Minerva User Guide

Run an HFT Collection

Start a High-Frequency Telemetry (HFT) collection from the UI or the HFT API, check its state, and stop it early if you need to. Before you start, prepare the switches and register their credentials. See Prepare Switches for HFT and Register Switch Credentials for HFT.

Start a Collection

  1. In the upper-right corner, click Settings menu and select Settings, and then select High Frequency Telemetry.

  2. Click New collection. The New high-frequency collection window opens. Click Next to move from one step to the next.

  3. On the Ports step, select the switches to sample. To find switches, search by device name, pod, SU, or rack. When you select a switch, all of its breakout interfaces are selected and listed under it; clear any you don't want. Only breakout interfaces, for example swp1s0, are listed. Interfaces that are oper down, admin down, or admin disabled are left out. To show or hide a switch's interfaces, click View or Hide. Only switches with registered credentials are listed. You do not need to know which collector manages each switch; NVIDIA Minerva sends the configuration through the right one.

  4. On the Metrics step, select up to 20 counters. Every counter you select is collected on every port you selected. When you select a scoped counter, a list of switch priorities, traffic classes, or priority groups appears below the counters. Select the ones to collect, or select All. The bottom of the window shows the number of ports and expanded counters. Each per-port counter counts once, and each scoped counter counts once for every priority, class, or group you select.

  5. On the Interval & duration step, choose a sampling interval and a duration, or enter your own values. You can enter a custom sampling interval in microseconds or milliseconds, from 100 microseconds to 12.75 milliseconds; it is rounded up to the next 50 microseconds. The duration can be from 1 second to 1 hour. You can enter it in seconds or minutes, for example 90s or 2m 30s. The Collection budget shows the collection's data rate, the estimated number of samples and data volume, and whether the collection is within the capacity limits.

  6. On the Review step, check the collection, and then click Start collection.

The configuration takes about 30 seconds to reach the switches, and then the collection state changes to Running. When the duration ends, or when you stop the collection, NVIDIA Minerva returns each switch to its configuration from before the collection.

A switch can run only one HFT collection at a time.

Collection States

The High Frequency Telemetry page lists active and completed collections. Select a collection to see its settings, its sessions (one per switch), and the state of each port, for example Collecting or Finished. While a collection runs, the Capacity & Limits panel shows the data rate it uses. When the collection ends, it moves from the Active list to the Completed list, and its data rate is freed up for new collections.

State

Meaning

Queued

NVIDIA Minerva is configuring the switches.

Running

At least one switch is still collecting. Some switches might already have failed; check each switch's session.

Finished

Every switch collected for the full duration.

Partial

Every switch has stopped; some finished and some failed.

Failed

Every switch failed.

Stopped

A user stopped the collection early.

Expired

The collection ended longer ago than the raw-sample retention period. Its raw samples are no longer available, and Explore is not shown.

Each port shows its own state: Queued, Collecting, Partial, Finished, Failed, or Stopped.

One switch failing doesn't stop the others. If some switches fail and others finish, the collection ends as Partial. See Troubleshoot HFT.

Available Counters

Scope

Counters

Per port

rx-byte, tx-byte, rx-packet, tx-packet, if-in-discards, pause-mac-ctrl-frames-rx, pause-mac-ctrl-frames-tx, ar-grade-high, ar-grade-low

Egress traffic class (0–15)

tc-occupancy, tc-watermark, tc-ecn-marked, tc-octets-tx

Switch priority (0–7)

sp-octets-rx, sp-pause-rx, sp-pause-tx

Ingress priority group (0–7)

pg-occupancy, pg-watermark, pg-headroom-occupancy, pg-headroom-watermark

Stop a Collection

Select the collection in the Active list, and then click Stop collection. NVIDIA Minerva removes the HFT configuration from each switch, and the collection state changes to Stopped.

Start a Collection with the API

The HFT API is at https://<platform-ip>/api/hft-control/v1 and uses the NVIDIA Minerva API key. See Configure an API Key.

curl -sk -X POST -H "X-API-Key: <api-key>" -H "Content-Type: application/json" \
  https://<platform-ip>/api/hft-control/v1 \
  -d '{
    "targets": [{"device": "leaf-01", "interfaces": ["swp1s0", "swp1s1"]}],
    "counters": ["rx-byte", "tx-byte"],
    "sample_interval_usec": 500,
    "duration_sec": 60
  }'

Field

Required

Description

targets

Yes

Switches and interfaces. An empty interfaces list means every breakout interface on the switch.

counters

Yes

Counter names from the Available Counters table

sample_interval_usec

Yes

Sampling interval in microseconds, from 100 to 12750, in steps of 50

duration_sec

Yes

Collection length in seconds

egress_tcs

For traffic-class counters

Traffic classes, 0–15

switch_priorities

For switch-priority counters

Switch priorities, 0–7

ingress_pgs

For priority-group counters

Priority groups, 0–7

The response includes the collection id. Use GET /<id> to check the collection, DELETE /<id> to stop it early, and POST /validate with the same body to check a request without starting it. A collection that names a switch without credentials is refused with no HFT credentials for switch.

Last updated: