Run the O-RAN Software Community hybrid M-Plane test runner (it-tifg) against a commercial O-RAN radio unit, the Benetel RAN650, while a real O-DU (Duranta, the COSMOS OpenAirInterface build) runs a live 5G cell on it. The O-RU then has two NETCONF managers at once: the O-DU over SSH, and the O-RU Controller of a full O-RAN SC SMO that the O-RU calls home to over TLS. Twelve test cases check the Call Home, reads, writes, alarms, software inventory and the M-Plane version, and then what is particular to hybrid mode: both sessions, the two roles, a controller change seen by the O-DU side, and an O-DU left untouched when the controller reconnects. The run puts the O-RU back exactly as it found it.
The simulator tutorial runs the whole hybrid M-Plane chain on one node with a simulated O-RU. This tutorial replaces both simulators with hardware:
smo.ndz: Kubernetes, SDN-R as the O-RU Controller), with it-tifg and the COSMOS test cases preinstalled. The O-RAN SC OAM controller in Docker (hybrid-mplane.ndz, the simulator tutorial's image) still works with hmp_controller: oam, with the original four test cases.The hybrid test cases (1xx) are the reason for the real O-DU: the O-RU must report two NETCONF sessions, one from the O-DU and one from the controller, and a change or a reconnect on the controller side must not disturb the O-DU. A test that writes to a shared radio must not leave a trace, so the run takes a snapshot of the O-RU's NETCONF configuration first and restores it at the end, also when a test fails.
| Test | What it checks | Based on |
|---|---|---|
| TC-001 | The O-RU's TLS Call Home reaches the controller and the handshake completes | O-RAN WG4 M-Plane conformance 3.1.1.5 |
| TC-002 | Reads: everything, a filtered subtree, configuration only | 3.1.4.1 / 3.1.4.2 |
| TC-004 | A write of Call Home timer leaves is stored (only the leaves the O-RU's YANG defines; a missing one is a WARN with the reason) | 3.1.10.1 |
| TC-005 | SSH Call Home (SKIP when the run uses TLS) | 3.1.1.1 |
| TC-006 | An out-of-range write is refused and the stored value is unchanged | 3.1.10.2 |
| TC-007 | The active alarm list can be read | 3.1.5.2 |
| TC-008 | The software inventory can be read (read-only) | 3.1.6 / 3.1.7 |
| TC-009 | The declared M-Plane version and the core O-RAN module set | 3.1.1 |
| TC-101 | Two sessions at once: the O-DU's and the controller's | hybrid |
| TC-102 | The two managers' accounts and NACM groups; with separate accounts, the O-RU must refuse the O-DU side an operation its own access rules deny | hybrid, related to 3.1.8.1 |
| TC-103 | A change made by the controller is notified to the O-DU side (netconf-config-change) |
hybrid |
| TC-104 | The controller's session is ended and comes back by Call Home; the O-DU's session and every carrier stay as they were | hybrid |
The 0xx tests follow O-RAN WG4 M-Plane conformance clauses but are not the conformance tests themselves. TC-103 and TC-104 need the O-DU's view, which Duranta's M-Plane client cannot report, so it-tifg opens its own sessions as the O-RU's O-DU account (hmp_oru_odu_user). The Benetel has one operator account, so on it TC-102 reports a shared account as WARN.
After completing this tutorial you will be able to:
it-tifg test cases against real hardware and read the O-RAN test-results JSON.| Difficulty | Advanced |
|---|---|
| Estimated time | About 20 min to load the two images (the SMO image is 23 GB) and 3 min for the SMO to start, then about 8 min per run (6 min of it is the O-DU bringing up the cell) |
| Domain / sandbox | sb1, osc or rrail: an O-RAN 7.2 fronthaul server for the O-DU, and an SMO server that reaches the O-RU's management address (it may be in another domain). This page uses srv1-lg1.sb1 (O-DU) and srv15.osc (SMO) |
| Images | duranta.ndz on the O-DU server, smo.ndz on the SMO node |
| Transmits? | Yes. The Benetel radiates a 5G cell on your O-RU reservation's channel while the run lasts |
| Tutorial package | cosmos-experiments-hybrid-mplane (cosmos-tutorial run hybrid-mplane-du) |
| Last verified | 2026-11-21: manual mode and teardown on srv1-lg1.sb1 + srv15.osc (smo-20261121b) with the Benetel RAN650 |
Background knowledge
it-tifg and its test cases. Do it first.smo.ndz runs and how it starts.Account & access
Devices / nodes
| Resource | Role | Qty | Notes |
|---|---|---|---|
O-RAN 7.2 fronthaul server, e.g. srv1-lg1.sb1 |
Duranta O-DU + OAI CN5G core | 1 | Intel E810 on the fronthaul fabric, PTP |
A server meeting the SMO's upstream minimum (64 GB of RAM, 20 vCPU), e.g. srv15.osc |
SMO: O-RU Controller + it-tifg |
1 | Reaches the O-RU's management address over its control network; needs internet access at boot (see the SMO page) |
| Benetel RAN650 | O-RU under test | 1 | Booked in the O-RU scheduler; shares one RF space with the LiteOn |
Disk images
| Image | Load onto | Provides |
|---|---|---|
duranta.ndz |
the O-DU server | Duranta, OAI CN5G, the fronthaul host layer, cosmos-oru-tools |
smo.ndz |
the SMO node | the full O-RAN SC SMO (Kubernetes 1.32, SDN-R), it-tifg with the COSMOS test cases, and the real-O-RU Call Home helpers |
Software components
| Component | Version | Source |
|---|---|---|
O-RAN SC SMO (SDN-R onap/sdnc-image 2.6.1 on OpenDaylight netconf 5.0.7) |
O-RAN SC SMO deployment | github.com/o-ran-sc/it-dep |
it-tifg hybrid M-Plane test runner |
ee6e413 + the COSMOS patches (TC-004 partial writes, TC-005..009, TC-101..104) |
github.com/o-ran-sc/it-tifg |
| Duranta O-DU | the build in duranta.ndz |
Duranta on COSMOS |
| Benetel RAN650 software | V2.1.0 (netopeer2 NETCONF server) | on the O-RU |
Spectrum / RF / special: the Benetel transmits on your O-RU reservation's channel (band n78 on this page). Only one of the Benetel and the LiteOn may be on a given channel.
console of your domain (cosmos-tutorial run hybrid-mplane-du)
│ ssh root@<du-node>, root@<smo-node>
▼
srv1-lg1 (duranta.ndz) srv15 (smo.ndz: SMO in Kubernetes)
┌──────────────────────────────┐ ┌──────────────────────────────────────────┐
│ Duranta O-DU + OAI CN5G core │ │ it-tifg ── RESTCONF ──► O-RU Controller │
│ │ 7.2 fronthaul (C/U-plane)│ │ │ O-DU-side client (SDN-R) │
│ │ NETCONF over SSH :830 ───┼──┐ │ ▲ TLS :4335 │
└───┼──────────────────────────┘ │ └───────────────────────────┼──────────────┘
▼ ▼ │ Call Home
┌─────────────────────────────────────────────────────────────────────────┴─────┐
│ Benetel RAN650 (netopeer2): session 1 = the O-DU (SSH), session 2 = the │
│ controller (TLS, opened BY the O-RU) │
└───────────────────────────────────────────────────────────────────────────────┘
The O-DU reaches the O-RU over the fronthaul fabric. The controller takes the Call Home on the SMO node's control network address: M-Plane traffic is light, so it does not need a data-plane interface. The O-DU-side test client (for TC-102 to TC-104) runs inside it-tifg on the SMO node and logs in to the O-RU over SSH as the O-RU's O-DU account.
# on console.sb1
omf stat -t benetel-oru.instrument.orbit-lab.org,liteon-oru.instrument.orbit-lab.org
omf load -i duranta.ndz -t srv1-lg1 -r 0 -o 2400 -b 600
omf tell -a on -t srv1-lg1
# on console.osc
omf load -i smo.ndz -t srv15 -r 0 -o 2400 -b 600
omf tell -a on -t srv15
Read the last line of each load: 1 node successfully imaged is success. If a server does not answer ping a few minutes after on, omf tell -a reset it once. The SMO starts by itself at boot (smo-up.service, about 3 minutes); the run waits for it.printf 'hmp_du_ru_console: console.rrail.cosmos-lab.org\n' > my.yml
export OMF_NODES=srv1-lg1.sb1.cosmos-lab.org,srv15.osc.cosmos-lab.org # the O-DU server first, then the SMO node
cosmos-tutorial check hybrid-mplane-du
cosmos-tutorial run hybrid-mplane-du --mode manual --params my.yml
check verifies both nodes and the console before anything changes (output from console.sb1, 2026-11-21, shortened: each probe runs on both nodes):
## check 'hybrid-mplane-du'
PASS subdomain sb1 is supported
PASS subdomain osc is supported
PASS osc's nodes are reached through console.osc.cosmos-lab.org
INFO srv1-lg1.sb1.cosmos-lab.org: image duranta.ndz (a hint only; the probes decide)
PASS srv1-lg1.sb1.cosmos-lab.org: O-DU server has an Intel E810 for the fronthaul (or is the SMO node)
PASS srv1-lg1.sb1.cosmos-lab.org: O-DU server has cosmos-oru-tools listing the Benetel (or is the SMO node)
INFO srv15.osc.cosmos-lab.org: image smo-20261121b.ndz (a hint only; the probes decide)
PASS srv15.osc.cosmos-lab.org: SMO node has the real-O-RU Call Home pieces (smo.ndz 20261121b or later, or hybrid-mplane.ndz 20261119b or later) (or is the O-DU server)
PASS srv15.osc.cosmos-lab.org: SMO node with smo.ndz has the controller file and the hybrid test cases (or is not an smo.ndz node)
PASS srv15.osc.cosmos-lab.org: SMO node has the RAM its controller needs, 64 GB for smo.ndz, 12 GB for hybrid-mplane.ndz (or is the O-DU server)
...
PASS the cosmos-experiments-duranta package on this console (the O-DU side is its duranta-72 tutorial, cell only: duranta_ue_mode none)
PASS the O-RU logins on this console - package cosmos-oru-credentials (/etc/cosmos/oru-credentials and /etc/cosmos/benetel_id_ed25519). ...
RUNNABLE
The run, 23:05 to 23:13 UTC (8 min):
## 0. Nodes, images, the O-RU
srv1-lg1.sb1.cosmos-lab.org: duranta.ndz
srv15.osc.cosmos-lab.org: smo-20261121b.ndz
O-RU benetel at <ru-address>; the controller takes its Call Home on <smo-address>:4335
## 1. SMO node: wait for the boot-time bring-up (smo-up.service)
[smo-callhome 22:54:00Z] TLS Call Home: <smo-address>:4335 and <smo-address>:30335 SSH Call Home: <smo-address>:4334 and <smo-address>:30266 RESTCONF: http://<smo-address>:30267/rests
## 2. O-DU: the duranta-72 tutorial, cell only
## 1. COLD power-cycle the O-RU (omf tell needs your ACTIVE O-RU reservation for this console's domain)
RU-UP after 70 s
## 4. O-DU bring-up (benetel-bringup.sh: channel, the O-RU's own state, VFs, IRQ pinning, hugepages, PTP, core, DU)
GOLDEN-PERSISTENT-OK
[23:11:00.208859] 00014935 [NR_RRC] I cell PLMN 001.01 Cell ID 2 is in service
fh: [23:12:52.401269] 0001493b [HW] I [o_du0][drop errors 0 ecpri errors 0 cp errors 0 up errors 0 pusch errors 0
cell in service
## 3. SMO node: the O-RU logins and this tutorial's helper
O-RU logins pushed to srv15.osc.cosmos-lab.org:/run/cosmos (tmpfs)
O-DU-side test client: oranbenetel@<ru-address>
## 4. TLS identity for the O-RU, into its keystore; the controller allows it
TLS material for benetel-ran650 (Call Home to <smo-address>:4335) in /var/tmp/cosmos/hybrid-mplane-du/tls
O-RU snapshot: 6 files
O-RU keystore + truststore: OK
controller: POST operations/netconf-keystore:add-trusted-certificate -> 204
controller: PUT data/odl-netconf-callhome-server:netconf-callhome-server/allowed-devices/device=benetel-ra -> 201
## 5. Run the it-tifg test cases (001 002 004 005 006 007 008 009 101 102 103 104) against the O-RU, Call Home triggered by the runner
overall: WARN testId 2cca9393-6c92-4e96-ab24-87fcd811e8ba
TC-001 PASS Transport and Handshake in IPv4/TLS Environment
TC-002 PASS Read-Only Data Retrieval (Unfiltered, Filtered, and Config-Only)
TC-004 PASS O-RU Configurability Test (Positive Case)
TC-005 SKIP M-Plane Transport over SSH Call Home (WG4 CONF 3.1.1.1, IPv4 positive)
TC-006 PASS O-RU Configurability Test (Negative Case) (WG4 CONF 3.1.10.2)
TC-007 PASS Retrieval of Active Alarm List (WG4 CONF 3.1.5.2)
TC-008 PASS Software Inventory Retrieval (WG4 CONF 3.1.6/3.1.7, read-only)
TC-009 PASS M-Plane Version and O-RAN Module Set (WG4 CONF 3.1.1 capability exchange)
TC-101 PASS NETCONF Sessions Verification (hybrid: O-DU and O-RU Controller)
TC-102 WARN Role-Based Access of the O-DU and the O-RU Controller (hybrid; WG4 CONF 3.1.8.1)
Shared account: the O-DU side and the O-RU Controller both log in as oranbenetel: hybrid role separation is not exercised on this rig
TC-103 PASS Cross-Client Change Visibility (hybrid: O-RU Controller change seen by the O-DU side)
TC-104 PASS Session Isolation (hybrid: O-RU Controller reconnect leaves the O-DU untouched)
## 6. Restore the O-RU (TC-004's write, the call-home client, keystore, truststore) and the controller
TC-004 leaf re-call-home-no-ssh-timer: DELETE 204
TC-004 leaf max-call-home-attempts: DELETE 500
ietf-netconf-server: running same, startup same
ietf-keystore: running same, startup same
ietf-truststore: running same, startup same
test material in the O-RU's persisted files: 0
RESTORE-OK
## 7. Fetch the results
TC-101 saw NETCONF sessions from: <du-address> <smo-address>
## 8. Build the MANUAL-mode result contract
overall WARN: TC-001 PASS, TC-002 PASS, TC-004 PASS, TC-005 SKIP, TC-006 PASS, TC-007 PASS, TC-008 PASS, TC-009 PASS, TC-101 PASS, TC-102 WARN, TC-103 PASS, TC-104 PASS
TC-102 WARN: see the notes in results/results_2cca9393-6c92-4e96-ab24-87fcd811e8ba_20261121T231227Z.json
The cell is still up. Tear it down with: cosmos-tutorial run hybrid-mplane-du --mode teardown
The run counts as passed (exit 0): WARN means a test ran and found a limitation, here the Benetel's single operator account (set hmp_accept_warn: false to count only PASS). TC-005 is skipped because the run uses TLS Call Home. max-call-home-attempts: DELETE 500: the controller could not delete the value TC-004 wrote (10). It is not saved in the O-RU's startup configuration: after the O-RU's next cold start neither TC-004 leaf is present (checked 2026-11-21).
The cell stays up after the run, so you can look at both sessions yourself (next section). Tear it down when you are done (see Cleanup).
The package's hybrid-mplane-du_manual.sh is the manual mode: every command it runs is printed with its step number, in the order above. Read it on the console in /usr/share/cosmos-experiments/hybrid-mplane-du/. The O-RU helper it ships to the SMO node, host/hmp-oru-callhome.sh, has three verbs: apply (snapshot the O-RU, install the TLS identity, allow it on the controller), callhome (add the Call Home client, the trigger it-tifg runs after it records the existing sessions) and restore. On smo.ndz the helper and cosmos-hmp-run find the controller (its RESTCONF address, the admin login file and the SMO's own address) in /etc/cosmos/hmp-controller.env, which the SMO writes at every boot.
Parameters you may change in an overlay (cosmos-tutorial params hybrid-mplane-du lists them all): hmp_tests (which test cases), hmp_controller (smo or oam), hmp_odu_client (the O-DU-side test client for TC-102 to TC-104), hmp_accept_warn, and hmp_du_ru_console.
hmp-oru-tls.sh on the SMO node makes a CA, an O-RU certificate (CN benetel-ran650) and a controller client certificate. The helper snapshots the O-RU's ietf-netconf-server, ietf-keystore and ietf-truststore datastores, then loads the O-RU key and the CA into them; the controller gets the CA in its trusted certificates and benetel-ran650 in its allowed Call Home devices.it-tifg records the O-RU's existing connections, then runs the trigger (HMP_CALLHOME_TRIGGER, a COSMOS patch to it-tifg), which adds the O-RU's Call Home client for <smo-address>:4335. netopeer2 connects within seconds; the controller creates the mount point benetel-ran650.get-schema) and writes only the leaves it defines.hmp_oru_odu_user, with the Benetel key). TC-103 subscribes to the O-RU's notifications there before the controller writes; TC-104 ends the controller's session and checks from the O-DU side that the O-DU's session and every carrier are unchanged while the O-RU calls home again. The controller's own session is identified by the SMO's address and its TLS transport, never by elimination, so the O-DU's session cannot be ended by mistake.trap, so also after a failure): TC-004's leaves are deleted, the Call Home client is removed, the three datastores are put back from the snapshot and compared, running and startup, and the controller forgets the device. RESTORE-OK is printed only when everything matches; otherwise the snapshot is kept and the run exits non-zero.cell in service from the O-DU, with zero fronthaul errors.overall: WARN on the Benetel: every test PASS except TC-102 (WARN, the shared account) and TC-005 (SKIP, the run uses TLS). The run's exit status is 0.TC-101 saw NETCONF sessions from: two addresses: the O-DU's and the SMO node's.After metric: O-DU-side sessions lost: [], Carriers changed: {} and the controller back within seconds.RESTORE-OK, with every datastore running same, startup same and test material in the O-RU's persisted files: 0.The run's result contract and the it-tifg results (results_<testId>_<time>.json, the RESTCONF debug log) are in the run directory that cosmos-tutorial prints, under results/. The fronthaul counters keep running in the O-DU log while the cell is up; zero drop, ecpri, cp, up and pusch errors after TC-004 show that the M-Plane test did not disturb the cell.
cosmos-tutorial run hybrid-mplane-du --mode teardown
The teardown restores the O-RU if a run left its snapshot, stops the O-DU, the core and the run's PTP daemons, and removes the O-RU logins and the tutorial's files from both nodes:
PLAY RECAP *********************************************************************
srv1-lg1.sb1.cosmos-lab.org : ok=6 changed=3 unreachable=0 failed=0 skipped=0 rescued=0 ignored=0
srv15.osc.cosmos-lab.org : ok=3 changed=1 unreachable=0 failed=0 skipped=0 rescued=0 ignored=0
Then power the Benetel off from your O-RU reservation's console, and the nodes:
omf tell -a offh -t benetel-oru.instrument.orbit-lab.org
omf tell -a offh -t srv1-lg1 # on console.sb1
omf tell -a offh -t srv15 # on console.osc
| Symptom | Cause | Fix |
|---|---|---|
Your reservation drives benetel-oru from rrail |
the O-RU booking belongs to another domain's console | set hmp_du_ru_console to that console in your overlay |
the O-DU cell did not come up |
the Duranta bring-up failed | read du/console.txt in the run directory; see the Duranta 7.2 tutorial |
smo-up on <node> is 'activating' for minutes |
the SMO is still starting (Kubernetes and about 80 pods) | wait; /var/log/smo-up.log on the SMO node. The SMO needs internet access at boot |
<node> has no smo-up.service |
the SMO node runs another image | load smo.ndz, or set hmp_controller: oam for hybrid-mplane.ndz |
TC-102 WARN: Shared account |
the O-DU and the controller log in as the same O-RU account | expected on the Benetel (one operator account). TC-102 passes only on an O-RU with separate O-DU and controller accounts |
The teardown reports the O-DU server UNREACHABLE |
ssh to the O-DU server can stall while the cell runs (heavy real-time load) | run the teardown again; it stops the DU the second time |
cosmos-oru-creds-push <node> failed: copying by scp |
the console's push tool refuses a node whose image was just loaded (new host key) | nothing to do: the run copies the logins itself |
refuses: a snapshot exists at step 4 |
an earlier run did not finish its restore | run the teardown, then run again |
RESTORE-DIFF, exit 4 |
the O-RU's datastores differ from the snapshot | the snapshot is kept on the O-RU; run the teardown, and if it still differs ask the testbed administrators |
A server does not come back after omf load |
it stayed off or hung at boot | omf tell -a on, then once omf tell -a reset |
!! live RU differs from the golden AFTER the DU configured it, with register values that belong to the next register in the list and have None entries, while the cell is in service with zero fronthaul errors |
some of the O-DU server's ssh reads of the O-RU failed while the cell was running, and the golden check counts a failed read as drift. It happened in the 2026-11-19 runs of this tutorial; replaying them against the stored O-RU configuration showed no real difference | not a real drift: the test results stand. For a fresh check, run cosmos-oru diff benetel duranta on the O-DU server before the teardown, while the cell is still up |
it-tifg: github.com/o-ran-sc/it-tifgAuthor(s): COSMOS team · Last verified: 2026-11-21 (manual mode and teardown, srv1-lg1.sb1 + srv15.osc, Benetel RAN650) · Tested image/release: duranta.ndz = duranta-w44-20261120b, smo.ndz = smo-20261121b · Tags: oran, m-plane, hybrid, netconf, call-home, tls, nacm, benetel, duranta, smo, it-tifg