Faster Bare-Metal Boots in OpenShift PoCs

Cut POST wait time on Dell, HPE, Cisco, and Lenovo servers during OpenShift bare-metal PoC redeploys, then turn memory checks back on before handback.

Posted Jul 31, 2026 Updated Aug 21, 2026

By Thomas Hall

12 min read

Personal site note: views expressed here are my own and do not necessarily represent Red Hat or any other organization.

PoC only. Disabling firmware memory tests hides failing DIMMs. Use this for short-lived labs, demos, and partner Proof of Concepts—not production clusters, regulated environments, or hardware you have not validated.

OpenShift bare-metal PoCs punish you twice: once while the installer waits on POST, and again every time you wipe a node and retry. On dense memory servers, software memory tests during UEFI/POST can add minutes per reboot. Across a multi-node rack and a week of agent-install / Metal3 / MachineConfig loops, that tax becomes the loudest part of the engagement.

This post is a field runbook for temporary OpenShift PoCs: what “memory checking” usually means, when it is acceptable to turn it off, vendor steps to disable and re-enable it on Dell, HPE, Cisco, and Lenovo, and a handback checklist before you return the gear or promote anything toward production.

Why PoC boot time hurts more than production

In production you reboot rarely and you want every hardware safety net. In a PoC you reboot constantly:

Assisted Installer / Agent-based Installer discovery boots
Ironic / Metal3 inspection and provisioning cycles
“Break the MachineConfig → reboot → watch the node come back” debugging
Day-2 operator installs that bounce workers
Full cluster wipe and rebuild between customer scenarios

Saving even five to fifteen minutes per cold boot compounds quickly. A six-node PoC that reboots twice a day for five days is sixty POST cycles. Cut ten minutes each and you reclaim ten hours—more than a full workday—of waiting.

What “memory checking” means during POST

Vendors name it differently, but the idea is the same: during power-on or warm reset, firmware spends time validating DIMMs before handing control to the boot device (PXE, ISO, disk, or virtual media).

Common categories:

Category	What it does	PoC impact
Software memory test	BIOS/UEFI walks or patterns system RAM	Often the big win to disable
Extended / enhanced memory test	Deeper DIMM validation; maps out bad ranks	Can add large, memory-size-scaled wait
Chipset MBIST / hardware BIST	Built-in self-test from the memory controller	May still run even if software test is off
Memory fast training / quick POST	Shorter training path on subsequent boots	Complementary; verify per generation

Dell documents that System Memory Testing is a software BIOS test and is distinct from chipset MBIST, which can still run every boot. Expect “faster,” not “instant.”

Tradeoffs for PoCs

Keep checks enabled	Disable for the PoC window
Unknown or freshly racked loaner hardware	Known-good lab gear with recent successful burn-in
Customer wants prod-like validation	Iteration speed is the engagement bottleneck
You already saw ECC / IML / SEL memory events	No memory faults in BMC/OS logs so far
Hardware will become the production cluster	Pure throwaway demo nodes

If correctable or uncorrectable memory errors appear after you disable tests, re-enable immediately and stop treating the node as healthy PoC inventory.

When to disable in a PoC

Disable (or confirm already disabled) when all of these are true:

The cluster is explicitly a PoC, lab, or demo.
Nodes will be wiped or returned at the end of the engagement.
You accept the risk of a latent DIMM fault showing up as random crashes instead of a clean POST failure.
You have a teardown step on the calendar to restore vendor defaults.

Keep checks on (or run one full test first) when:

The servers are production candidates the customer will keep
DIMMs were just reseated, upgraded, or moved between systems
The customer’s success criteria include hardware validation
You are in a regulated or air-gapped demo where unexplained instability is worse than a slow boot

Vendor runbooks

Menu labels and Redfish attribute names drift by generation. Treat the names below as the common ones; confirm on your firmware before you script a rack.

Dell PowerEdge (iDRAC / System BIOS)

Setting: System Memory Testing
Attribute (common): BIOS.MemSettings.MemTest
Values: Enabled / Disabled
Generations: Widely present on 14G/15G/16G-class PowerEdge with iDRAC9; verify with racadm get BIOS.MemSettings.MemTest.

Console

Reboot and enter System Setup (F2), or open iDRAC → Configuration → BIOS Settings.
Open Memory Settings.
Set System Memory Testing to Disabled.
Save and reboot.

RACADM (disable for PoC)

  
racadm -r <idrac-host> -u <user> -p <password> \
  set BIOS.MemSettings.MemTest Disabled

racadm -r <idrac-host> -u <user> -p <password> \
  jobqueue create BIOS.Setup.1-1 -r pwrcycle -s TIME_NOW

RACADM (re-enable before handback)

  
racadm -r <idrac-host> -u <user> -p <password> \
  set BIOS.MemSettings.MemTest Enabled

racadm -r <idrac-host> -u <user> -p <password> \
  jobqueue create BIOS.Setup.1-1 -r pwrcycle -s TIME_NOW

Dell’s software memory test is separate from chipset MBIST. Disabling MemTest shortens POST; it does not remove every hardware self-test.

HPE ProLiant (iLO / UEFI System Utilities)

Setting: Extended Memory Test
Redfish attribute (common): ExtendedMemTest
Values: Enabled / Disabled
Generations: Documented on Gen10/Gen10 Plus ProLiant with iLO 5 and Gen11 ProLiant with iLO 6; confirm with a GET of /redfish/v1/systems/1/bios.

HPE’s own description: when enabled, the system validates memory during initialization, can map out failed DIMMs to the IML, and can significantly increase boot time. For PoCs, ensure it is Disabled (many images already default that way—verify; do not assume).

Related options worth checking on your generation (names vary):

Memory Fast Training — shorter training on warm boots
Boot-time optimization / POST options in System Utilities

Console

Enter UEFI System Utilities during POST (often F9), or use the iLO remote console.
Open System Configuration → BIOS/Platform Configuration (path varies).
Find Extended Memory Test and set it to Disabled.
Save and reboot.

Redfish (disable for PoC)

  
curl -k -u "<user>:<password>" -X PATCH \
  "https://<ilo-host>/redfish/v1/systems/1/bios/settings/" \
  -H "Content-Type: application/json" \
  -d '{"Attributes":{"ExtendedMemTest":"Disabled"}}'

Then reset the server so pending BIOS settings apply (iLO virtual power / Redfish Reset action).

Redfish (re-enable before handback)

  
curl -k -u "<user>:<password>" -X PATCH \
  "https://<ilo-host>/redfish/v1/systems/1/bios/settings/" \
  -H "Content-Type: application/json" \
  -d '{"Attributes":{"ExtendedMemTest":"Enabled"}}'

Reset again after the PATCH.

Cisco UCS (CIMC / UCS Manager / Intersight)

Setting: Enhanced Memory Test (older docs: Advanced Memory Test)
CIMC token (common): AdvancedMemTest
Values: Disabled / Enabled / Auto (platform default is often Auto)
Scope note: Cisco documents that enhanced testing applies to certain DIMM vendors/populations; always confirm on the exact C-Series or B-Series model.

Cisco’s guidance in UCS Manager docs: enhanced memory tests during boot increase boot time based on memory size. For PoCs, set the policy to Disabled unless you intentionally want validation on first rack-up.

UCS Manager / Intersight BIOS policy

Create or edit a BIOS policy used only by the PoC service profile / server profile.
Under memory-related tokens, set Enhanced Memory Test to Disabled.
Associate the policy and reboot the server(s) so the CIMC buffer applies.

Keep a second policy (or the customer’s baseline) with the original value for handback.

CIMC CLI sketch (standalone C-Series)

Exact subcommand paths vary by IMC release. Pattern:

scope bios
scope advanced
set AdvancedMemTest Disabled
commit

Reboot the server after commit. To restore:

set AdvancedMemTest Auto
commit

(or Enabled if that was the pre-PoC value—capture it before you change anything).

Lenovo ThinkSystem (XClarity / UEFI) — common in partner labs

Setting: Memory Test
OneCLI examples (Intel-class, common):
Memory.MemoryTest and IMM.UEFIMemoryTest
Values: Enable/Disable variants; defaults are often Enabled

Console

Enter UEFI Setup (F1 on many ThinkSystem models).
Open System Settings → Memory (label varies slightly by CPU family).
Set Memory Test to Disabled.
Save and reboot.

OneCLI (disable for PoC)

  
OneCli config set Memory.MemoryTest Disable \
  --bmc <user>:<password>@<xcc-host>

OneCli config set IMM.UEFIMemoryTest Disabled --override \
  --bmc <user>:<password>@<xcc-host>

Restart the system for the change to take effect. On XCC firmware older than d8e130f-2.60, use IMM.UEFIMemoryTestOptions instead of IMM.UEFIMemoryTest (some AMD EPYC platforms, such as SR645/SR665, use this path too); check Lenovo’s OneCLI memory-test topic for your exact firmware stream.

OneCLI (re-enable before handback)

  
OneCli config set Memory.MemoryTest Enable \
  --bmc <user>:<password>@<xcc-host>

OneCli config set IMM.UEFIMemoryTest Repair --override \
  --bmc <user>:<password>@<xcc-host>

(Use the restore values that match what you captured at engagement start; Lenovo’s “enable advanced test” flow is more nuanced than a single boolean.)

Supermicro and other lab vendors

Supermicro and whitebox boards often expose Memory Testing, Quick Boot, or vendor-specific POST options under Advanced → Chipset / Memory. There is less uniformity than Dell/HPE/Cisco. For PoCs:

Photograph or export the BIOS page before changes.
Prefer Quick Boot / skip memory test style options when present.
Do not script a rack until you have confirmed the Redfish/IPMI attribute on one golden node.

Automating a multi-node PoC rack

For more than a couple of nodes, do this once from a jump host with Ansible and Redfish rather than walking consoles. Keep credentials out of git; use environment variables or a local vault file that never leaves the laptop.

Example pattern with community.general.redfish_command / raw URI calls (attribute names differ by vendor—parameterize them):

  
# poc-bios-memory-test.yaml
---
- name: Set BIOS memory test for OpenShift PoC nodes
  hosts: bmc
  gather_facts: false
  vars:
    # dell_memtest | hpe_extended | cisco_enhanced — choose per inventory group
    redfish_base: "https://"
  tasks:
    - name: Stage HPE ExtendedMemTest Disabled
      when: vendor == "hpe"
      uri:
        url: "/redfish/v1/systems/1/bios/settings/"
        method: PATCH
        user: ""
        password: ""
        force_basic_auth: true
        validate_certs: false
        headers:
          Content-Type: application/json
        body_format: json
        body:
          Attributes:
            ExtendedMemTest: Disabled
        status_code: [200, 204]

    - name: Create Dell BIOS job for MemTest Disabled
      when: vendor == "dell"
      # Prefer dellemc.openmanage or racadm; sketch shown for clarity
      ansible.builtin.command:
        argv:
          - racadm
          - -r
          - ""
          - -u
          - ""
          - -p
          - ""
          - set
          - BIOS.MemSettings.MemTest
          - Disabled

Run Ansible from a project-local venv:

  
python3 -m venv .venv
source .venv/bin/activate
pip install ansible
ansible-galaxy collection install community.general dellemc.openmanage
ansible-playbook -i inventory.yaml poc-bios-memory-test.yaml

After staging BIOS settings, reboot the nodes once before you start the OpenShift install so the first discovery boot already benefits.

Validation during the PoC

After the change, confirm three things:

Setting stuck: re-read the attribute (racadm get, Redfish GET, UCS policy view). Pending vs current values are a common footgun.
POST is actually shorter: time chassis power-on to PXE/agent start on one node before and after.
No new memory faults: watch iDRAC LifeCycle / HPE IML / UCS SEL, and on booted nodes check dmesg / EDAC / MCE messages.

If a node starts crashing, dumping MCE, or logging uncorrectable ECC:

Remove it from the PoC schedulable set.
Re-enable the vendor memory test.
Run the vendor diagnostics or replace DIMMs before trusting it again.

PoC teardown and handback

Put this on the engagement checklist next to “delete cluster” and “revoke pull secrets”:

Restore System Memory Testing / Extended Memory Test / Enhanced Memory Test / Memory Test to the pre-PoC value (capture it on day one).
Reboot once and confirm the restored value is active, not merely pending.
Optionally run one vendor full memory diagnostic on loaner gear before shipping it back.
If the customer is keeping the hardware as a production seed, leave checks enabled and do not hand over a “PoC-tuned” BIOS as the baseline.

Recommended PoC policy

Situation	Policy
Red Hat / partner lab kit, known-good history	Disable software/extended tests for the PoC window
Customer loaners, unknown DIMM health	One full test on day one, then disable if clean
Prod-candidate hardware	Leave enabled; accept the boot tax
Any memory error during the engagement	Re-enable immediately; quarantine the node
Engagement end	Restore prior BIOS values before handback

Wrap-up

Firmware memory tests are the right default for servers that matter. They are the wrong default for a throwaway OpenShift PoC where you will reboot the same six nodes until the demo story works. Disable the vendor software or extended memory test on Dell, HPE, Cisco, and Lenovo to reclaim POST time; keep a restore path; and never confuse “PoC fast” with “production ready.”

This pairs well with edge form-factor and hosted vs virtualized control plane labs that lean on Agent-based Installer, Metal3, and Redfish-heavy reboot loops—use the PoC BIOS tweak for the engagement window, then restore before you call the gear production-ready. For the rest of a bare-metal OpenShift PoC (prerequisites through install), start at how to get started with an OpenShift PoC or the OpenShift PoC overview.

Faster Bare-Metal Boots in OpenShift PoCs

Why PoC boot time hurts more than production

What “memory checking” means during POST

Tradeoffs for PoCs

When to disable in a PoC

Vendor runbooks

Dell PowerEdge (iDRAC / System BIOS)

Console

RACADM (disable for PoC)

RACADM (re-enable before handback)

HPE ProLiant (iLO / UEFI System Utilities)

Console

Redfish (disable for PoC)

Redfish (re-enable before handback)

Cisco UCS (CIMC / UCS Manager / Intersight)

UCS Manager / Intersight BIOS policy

CIMC CLI sketch (standalone C-Series)

Lenovo ThinkSystem (XClarity / UEFI) — common in partner labs

Console

OneCLI (disable for PoC)

OneCLI (re-enable before handback)

Supermicro and other lab vendors

Automating a multi-node PoC rack

Validation during the PoC

PoC teardown and handback

Recommended PoC policy

Wrap-up

Further reading

Trending Tags

Why PoC boot time hurts more than production

What “memory checking” means during POST

Tradeoffs for PoCs

When to disable in a PoC

Vendor runbooks

Dell PowerEdge (iDRAC / System BIOS)

Console

RACADM (disable for PoC)

RACADM (re-enable before handback)

HPE ProLiant (iLO / UEFI System Utilities)

Console

Redfish (disable for PoC)

Redfish (re-enable before handback)

Cisco UCS (CIMC / UCS Manager / Intersight)

UCS Manager / Intersight BIOS policy

CIMC CLI sketch (standalone C-Series)

Lenovo ThinkSystem (XClarity / UEFI) — common in partner labs

Console

OneCLI (disable for PoC)

OneCLI (re-enable before handback)

Supermicro and other lab vendors

Automating a multi-node PoC rack

Validation during the PoC

PoC teardown and handback

Recommended PoC policy

Wrap-up

Related posts

Further reading

Trending Tags