Getting Started

Important

Start Here - New to Karios? Use this onboarding order:

1. Complete Quick Start Checklist first.

2. Use Getting Started (this page) for context, terminology, and prerequisites.

3. Continue to ISO Install for the 9-step graphical installer and platform image delivery.

4. After the installer completes and the node reboots, open Control Center, the Karios web UI, at the final HTTPS URL shown on the console.

5. Use K-Store to apply platform updates and enable the optional plugins your deployment needs.

This page does not replace the checklist. It explains the terms and platform context used in that first-step readiness check.

For shared term definitions, see Glossary or section 3.2 Key Terminology below.

1. Welcome to Karios

Welcome to Karios. This guide helps new users understand what the platform does and how to begin in the correct order.

2. What is Karios?

Karios is a unified infrastructure operating system that simplifies how organizations build and manage modern IT infrastructure. It brings together compute, virtualization, Kubernetes orchestration, security, security operations, observability, and power intelligence into a single hardware-agnostic platform, enabling teams to provision, operate, and optimize infrastructure across datacenters, edge environments, and hybrid deployments.

        flowchart LR
    Op[Operator] --> CC[Control Center<br/>Web UI]
    CC --> Core[Karios Platform<br/>Control Plane]
    Core --> Compute[Compute &<br/>Virtualization]
    Core --> Storage[Storage<br/>K-Store / DFS]
    Core --> Network[Networking<br/>VPCs / SDN]
    Core --> K8s[Kubernetes<br/>K-Forge]
    Core --> Sec[Security<br/>K-Shield]
    Core --> Obs[Observability<br/>K-Trace]
    Core --> BM[Bare-Metal<br/>Lifecycle / BMO]
    

2.1. Platform Scope at a Glance

Karios is not only a hypervisor layer. It covers the full infrastructure workflow:

• Karios OS + graphical installer for base host provisioning

• 9-step install wizard (License, Target Disk, Network & NIC, Administration, Location, Hardware, Summary, Install, Reboot) for first-time control plane setup

• Control Center UI for day-to-day infrastructure operations after the installer completes and the node reboots

• K-Store for platform updates and the optional plugins that expose the sections below - online, air-gapped, or by USB

• K-Forge for bare-metal host lifecycle and hardware operations (BMO)

• Karios DFS for distributed storage operations (Dashboard, Storage, Block Storage, Object Storage, Infrastructure, Monitoring, Advanced)

• Infrastructure, Compute, Network, Storage, Compute Policies, Observability, Kubernetes, Security, and Security Operations modules in one operational plane

3. What You’ll Learn

After this section, you should understand:

• The end-to-end onboarding path

• Minimum prerequisites before you start

• Where to go next in the docs

3.1. Full Documentation End-State

After completing the full documentation flow, a new operator should be able to:

• Install Karios on supported hardware and reach Control Center

• Configure users, roles, infrastructure, storage, networking, and compute policies

• Provision and operate VMs and Kubernetes clusters

• Run security scans, interpret diagnostics, and triage operational alerts

• Execute support escalation with complete evidence and clear impact context

3.2. Key Terminology

Terms used throughout this guide:

Term	Technical Meaning	Plain English
Zone	Top-level infrastructure boundary	A datacenter location or site
Pod	Network segment inside a zone	A group of servers on the same network block
Cluster	Logical group of hosts	Multiple physical servers operating together
Host	Physical compute node	An actual machine with CPU/RAM
BMC	Baseboard Management Controller	Remote power/console control interface
BMO	Bare Metal Orchestration	Taking a bare server through to a working Karios node, driven through its BMC
Instance Profile	VM compute sizing template	VM size definition
VM Template	Prebuilt virtual machine image	Ready-to-use operating system image
Boot Image	Bootable installation media	Installer disk image
CIDR	IP range notation	Network block format like 192.168.1.0/24
Virtual Router	Software router for isolated guest networks	Auto-created network gateway for VM traffic
Compute Policy	Reusable resource policy object	Predefined profile used when creating or operating resources
Volume Profile	VM disk performance/capacity profile	Storage tier template for VM volumes

4. Platform Overview

4.1. Key Features

• Zero-touch provisioning across virtually any hardware

• Unified virtualization and Kubernetes orchestration

• Integrated security, observability, and automation

• One platform, one operational model

4.2. Beyond Virtualization

Traditional infrastructure stacks are fragmented, expensive, and operationally brittle. Karios is built to remove those constraints and simplify operations across the full infrastructure lifecycle.

Whether you are modernizing legacy infrastructure or deploying at the edge for the first time, Karios reduces complexity, lowers cost, and improves visibility and control.

4.3. Core Philosophy

Karios follows three core principles:

• Simplicity: Make complex infrastructure operations predictable.

• Integrity: Keep infrastructure behavior consistent and defensible.

• Agility: Allow fast, controlled iteration as requirements evolve.

Operational model:

• Hardware-agnostic lifecycle workflows reduce vendor lock-in complexity.

• Unified operations reduce tool sprawl and handoff friction.

• Security and observability are treated as core operations, not add-ons.

Who Karios is built for:

• Enterprises modernizing virtualization and Kubernetes

• Organizations constrained by power, space, or cost

• Edge, telecom, and distributed infrastructure deployments

• Government, utilities, finance, healthcare, education, and smart-city environments

• Teams that want fewer tools and better outcomes

5. Infrastructure Deployment: Network and Hardware Requirements

Use this section to validate deployment prerequisites before onboarding infrastructure.

Important

For the full onboarding flow in this documentation (the 9-step graphical installer, then Control Center setup), satisfy the 5.1 Network Requirements and 5.2 Hardware Requirements gates first, then use at least the 5.3 Deployment Profiles and Sizing baseline for the profile you have chosen.

5.1. Network Requirements

5.1.1. VLANs - 4 Required

Network	Purpose	Notes
Management	Platform services, DNS	Required
Storage	Storage traffic	Required
Public	Public network	Required
Guest	VM tenant traffic	Required
OOB (network role)	Out-of-band management network	Optional. This installer role is not currently used and is left unselected. The BMC itself is still required - see the hardware requirements below.

5.1.2. VLAN Rules

• All VLANs must be unique with no duplicates.

• No CIDR overlaps are allowed between any networks.

• All IPs must be within their respective declared CIDR range.

• Karios uses static addressing. Node addresses are not assigned by DHCP.

• A reachable DNS server is required.

5.1.3. Management Network

• Usable IP range must be greater than 20 IPs because 20 IPs are reserved for VIPs.

• Minimum IP range is enforced at validation time.

5.1.4. Storage Network

• Usable IP range must be greater than 20 IPs because 15 VIPs and 5 IPs are reserved.

• Storage IP range must be greater than or equal to the management IP range. It can exceed it, but it cannot be smaller.

5.2. Hardware Requirements

5.2.1. Supported Manufacturers

• Supermicro

• Dell

• HPE

• Cisco

• Lenovo

• MegaRAC SP-X (MSI, ASRock Rack and other AMI-based boards)

See 6. Supported OEMs for the BMC licence each one requires and the procedure that puts its storage controller into pass-through.

5.2.2. Disk Controller Mode

The storage controller must present disks individually, in HBA / JBOD / pass-through mode. The installer refuses hardware-RAID volumes and cannot install onto a controller-backed RAID set.

Set the controller to HBA/JBOD in firmware before booting the installer. Changing controller mode destroys any existing hardware-RAID volume, so do this before there is anything on the machine worth keeping.

5.2.2.1. ZFS and RAID Controllers

The short version. ZFS is compatible with all major RAID controllers, but only when the controller is placed in Non-RAID (passthrough) mode. This is a settings change, not a hardware change. Do not swap the card.

ZFS must be given the physical disks directly. Any mode where the controller presents a virtual disk instead of the real drives will appear to work at first and will quietly remove the protections ZFS exists to provide. Before installing, put the controller into Non-RAID / HBA / JBOD mode. The exact name varies by vendor. See 6. Supported OEMs for the procedure on each supported manufacturer.

Why ZFS needs the raw disk. ZFS is not only a filesystem. It is a combined volume manager, RAID engine and filesystem, and it is built on one assumption: it owns the disks.

Every block ZFS writes carries a checksum. When it reads a block back and the checksum does not match, ZFS fetches a good copy from another disk, returns the correct data to the application, and rewrites the bad copy. That self-healing is the core of the product, and it only works if ZFS can see each physical disk, knows which one holds which copy, and can write to them independently.

A RAID controller in RAID mode breaks that assumption. It merges the drives into one virtual disk and hands that to the operating system. ZFS sees a single device where several exist.

5.2.2.2. Per-Manufacturer Controller Setup

The pass-through procedure differs on every manufacturer, and so does the BMC licence you must buy before Karios can mount its installer.

See 6. Supported OEMs for the per-manufacturer tabs covering Supermicro, Dell, HPE, Cisco, Lenovo and generic hardware.

5.2.3. Node Specifications

Per-role minimums. For recommended figures and AIO sizing, see 5.3. Deployment Profiles and Sizing.

Component	Management Server	Agent / Compute Node
CPU (logical processors)	12 minimum (Management-only); 16 (AIO)	12 minimum
Bare-metal CPU guidance	≥6C / 12T (Management-only); ≥8C / 16T (AIO)	≥6C / 12T
RAM	32 GB minimum	32 GB minimum
Disks (minimum)	1+ (Management-only) / 2+ (AIO, one left empty for Ceph)	3+ (excluding OS disk)
BMC (IPMI/iDRAC/iLO)	Required	Required
EFI Bootloader	Required	Required
CPU Virtualization (VT-x / AMD-V)	Required	Required

5.2.4. Agent / Compute Node Disk Requirements

• OS installation disk: Tier 1 or Tier 2, at the user’s choice.

• Remaining disks: 3 or more disks, excluding the OS disk.

• The remaining disks must include at least one Tier 1 disk, one Tier 2 disk, and one Tier 3 disk.

• Every disk the installer can use must be at least 32 GB and internally attached.

• USB and other removable devices, SAN/iSCSI/FC/FCoE LUNs, and the installer medium itself are not counted as usable disks.

5.2.5. Disk Tier Classification

Tier	DWPD Range	Classification
Tier 1	10+	High endurance (NVMe/enterprise SSD)
Tier 2	3 - 10	Mid endurance
Tier 3	1 - 3	Standard endurance

Choose the profile that matches your environment after the deployment prerequisites above are satisfied.

5.3. Deployment Profiles and Sizing

Karios installs in one of two shapes, and the shape decides the sizing. Choose the profile first, then size from its column in 5.3.1. Everything in this section applies to every manufacturer; nothing here changes by vendor.

	AIO (All-in-One)	Cluster (Management + Agents)
Intended use	Proof of concept, labs, evaluation	Production
High availability	None. Single control node.	Yes, with three control nodes behind VIPs
Pods	One only	Multiple
Network fabric	Single fabric	Shared (1 NIC) or Dedicated (2 NICs)
Licensing	Licensed per agent socket	Control nodes are free; licensed per agent socket

Danger

AIO is a proof-of-concept and laboratory topology. It is not a production topology. It runs the entire control plane on one node, so it has no high availability and cannot grow beyond a single pod. Losing that node loses the deployment.

5.3.1. Sizing by role

Resource	AIO - Minimum	AIO - Recommended	Management - Minimum	Management - Recommended	Agent - Minimum	Agent - Recommended
CPU / Logical Processors	16	24	12	16	12	16
Bare-metal CPU guidance	≥8C / 16T	≥12C / 24T	≥6C / 12T	≥8C / 16T	≥6C / 12T	≥8C / 16T
RAM	32 GB	64 GB	32 GB	64 GB	32 GB	32-64 GB
OS / Boot Disk	≥240 GB SSD/NVMe	≥480 GB SSD/NVMe	≥240 GB SSD/NVMe	≥480 GB SSD/NVMe	≥240 GB SSD/NVMe	≥480 GB SSD/NVMe
K-DFS / OSD disks	Separate disk(s) required when K-DFS enabled	Dedicated enterprise SSD/NVMe	N/A unless hosting storage	Dedicated if used	Separate disk(s) required when K-DFS enabled	Dedicated enterprise SSD/NVMe
Network	10 GbE	10/25 GbE	1 GbE	10 GbE	10 GbE	10/25 GbE
Architecture	x86-64	x86-64	x86-64	x86-64	x86-64	x86-64
Deployment target	Bare metal / supported VM	Bare metal preferred	Bare metal / supported VM	Bare metal preferred	Bare metal	Bare metal preferred

5.3.2. Prerequisites common to every profile

These apply to AIO and cluster alike, and to every manufacturer:

• EFI bootloader required. Legacy and CSM boot are refused by the installer.

• Secure Boot disabled in firmware, or plan to enrol the Karios key from the local or BMC console when prompted.

• CPU virtualization (VT-x or AMD-V) available and enabled in BIOS.

• BMC (IPMI, iDRAC, iLO or equivalent) reachable with validated credentials.

• Storage controller in pass-through, HBA or JBOD. This is the one prerequisite whose procedure differs by manufacturer. See 6. Supported OEMs.

• Static addressing. Karios does not use DHCP for node addresses, and a reachable DNS server is required.

5.4. Environment Sizing Calculator (Example)

Example workload target: 20 VMs (10 small, 8 medium, 2 large)

VM demand model:

• Small VM: 2 vCPU, 4 GB RAM, 50 GB disk

• Medium VM: 4 vCPU, 8 GB RAM, 100 GB disk

• Large VM: 8 vCPU, 16 GB RAM, 200 GB disk

Calculated total VM demand:

• CPU (vCPU): (10x2) + (8x4) + (2x8) = 68 vCPU

• RAM: (10x4) + (8x8) + (2x16) = 136 GB

• Storage: (10x50) + (8x100) + (2x200) = 1700 GB

Add platform/HA overhead:

• CPU (vCPU): 68 x 1.3 = 89 vCPU

• RAM: 136 x 1.3 = 177 GB

• Storage: 1700 x 1.5 = 2550 GB (snapshots/growth buffer)

Example physical allocation (target outcome), using the role figures in 5.3.1:

• Option A: 6 agents x 16 logical processors, 32 GB each (96 logical processors, 192 GB)

• Option B: 8 agents x 12 logical processors, 32 GB each (96 logical processors, 256 GB)

• Control plane: 1 management node, 12 logical processors, 32 GB, or three for high availability

• Storage: at least 3 K-DFS disks per agent (see 5.2.4). Option A: 18 x 960 GB = 17.3 TB raw (about 5.8 TB usable at 3x replication); Option B: 24 x 960 GB = 23 TB raw (about 7.7 TB usable). Keep raw usage below 80% (85% is the critical threshold); each node’s boot disk is separate.

Tip

For HA planning, ensure the environment can tolerate one-host failure while still meeting critical workload demand.

5.5. Hypervisor VM Prep (Lab/Nested Only)

Use this section only for lab or nested testing environments.

Warning

Supported production deployments should run on bare metal. Hypervisor-based deployments are for limited lab experimentation.

Why these settings matter:

• They reduce common virtualization compatibility issues during install and first boot.

• They avoid known performance bottlenecks from unsupported virtual controllers and NIC models.

• They provide a consistent baseline before troubleshooting cluster behavior.

5.5.1. ESXi 7.0 U2 or Greater

Setting	Value
Guest Family	Karios-compatible Linux family
Guest OS Version	Linux 6.x / Debian 12 compatible (64-bit)
Expose Hardware Assisted Virtualization to Guest OS	Yes
CPU	16 cores
Memory	Minimum 16 GB
Hard Disk Provisioning	Thin Provision
Disk Mode	Dependent
Controller	VMware Paravirtual SCSI controller. Do not use LSI Logic SAS.
Network Adapter	VMXNET3. Do not use E1000e because it is not 10GB capable.
VM Options	Firmware EFI and Secure Boot unchecked.

5.5.2. Proxmox VE 8 and 9

Area	Setting
OS	OS Type: Linux; Version: Linux 6.x compatible
System	Machine: Q35; BIOS: OVMF; Add EFI Disk: Checked; Pre-Enroll Keys: Unchecked; QEMU Agent: Checked
Disks	Bus/Device: VirtIO Block; Storage: Select the storage pool; Format: RAW (do not use Qcow2); Cache: No Cache; Discard: Checked; Async IO: io_uring or threads
CPU	Type: Host (16 cores)
Network	Model: VirtIO

5.5.3. Hyper-V

Area	Setting
Generation	Generation 2 only
Memory	Startup Memory: 16 GB; Use Dynamic Memory: Unchecked
Network	Connection: Gen 2 synthetic
Disks	Format: .vhdx; Type: Fixed
Post Creation - Security	Secure Boot: Unchecked
Post Creation - Network Adapter	Enable MAC Address Spoofing: Checked
Post Creation - Integration Services	Ensure all services are checked
Processor	4-16 vCPU (test for minimums)

5.6. Obtaining Installation Media

Alternative Download Methods

For environments with restricted internet access:

• Contact sales@karios.com for offline media delivery

• Request access to mirror sites for your region

Creating Bootable Media

USB Boot (Recommended for physical servers)

Use ISO Install Guide as the single step-by-step source for creating bootable media with Rufus, Etcher, or dd.

The ISO Install guide also contains the canonical USB media baseline, including minimum capacity, interface, speed guidance, and reliability notes.

Note

Keep this page as the onboarding summary and use the ISO Install guide as the canonical execution workflow.

IPMI Virtual Media (Recommended for remote installations)

1. Access your server’s BMC/IPMI interface

2. Navigate to Virtual Media or Remote Console

3. Mount ISO file from local machine or HTTP URL

4. Set boot order to Virtual CD/DVD

5. Reboot server

Download Size and Network Requirements

• ISO size: ~2-4 GB (version dependent)

• Karios platform image (.qcow2) size: ~5 GB - the template, bundled with the ISO in the same .zip download rather than fetched separately. It is delivered to the node from a second machine partway through the install

• Minimum download speed: 10 Mbps recommended

• Estimated download time: 5-30 minutes per artifact (depending on connection)

Version Selection Guidance

• Latest stable release: Recommended for new deployments

• Version-pinned release from the archive index: use when change-control requires fixed versioning

Tip

Bookmark the installer bundle: https://download.karios.com/installer/karios-installer.zip

6. Supported OEMs

Karios runs ZFS and Ceph directly on physical disks, so every storage controller must be put into pass-through, HBA or JBOD before the installer runs. The mode has a different name on every manufacturer, and reaching the screen that sets it takes a different key on every manufacturer too.

Two things differ by vendor: the BMC licence you must buy before Karios can mount its installer, and the pass-through procedure. Sizing does not vary by vendor and is covered in section 5 above.

Important

The Karios installer wipes every internal fixed disk in the chassis, whether you select it or not. Clearing a controller configuration also destroys every logical drive on it. Anything that must survive has to be physically removed first.

Note

Always prove pass-through from Linux, never from the BMC inventory page, which reports a MegaRAID controller as present and enabled in both RAID and JBOD mode. On an SSD the most reliable single check is lsblk -dD: a RAID virtual disk reports 0B discard because it hides TRIM, a passed-through drive reports a real value.

6.1 At a glance

Two decisions are made before a server is ordered, and one is made at the console. The table below answers all three for each manufacturer. Full procedures follow in the tabs.

Manufacturer	BMC licence to order	Controller to quote	Mode to set on a RAID card	Key at power-on
Supermicro	SFT-DCMS-SINGLE, one per node	-L8e suffix, or onboard SATA only	JBOD, controller then each drive	DEL
Dell	iDRAC9 Enterprise (iDRAC10 Enterprise on 17G)	HBA355i	Enhanced HBA, plus Non-RAID per drive	F2
HPE	iLO Advanced	E208i-a, or no controller at all	Controller Mode HBA	F10 or F9
Cisco	none, Cisco IMC is included	UCSC-SAS- part number	JBOD, controller then each drive	F2, or no reboot at all
Lenovo	XCC Enterprise / Platinum / Premier by generation	440-8i or 440-16i	JBOD (Broadcom) or HBA Mode (Microchip)	F1

Tip

Quoting a controller from the middle column removes the entire procedure. Those parts have no RAID mode to disable, so there is nothing to configure, nothing to get wrong on site and nothing to verify beyond a single lsblk. Where an existing fleet already carries a RAID card, the tabs below cover it.

6.2 Per-manufacturer procedures

Supermicro

BMC licence required

SFT-DCMS-SINGLE, one per node, X11 onward

Supermicro gates virtual media, BIOS configuration and BMC firmware update behind a licence key. Without it Karios cannot mount its installer ISO.

Check with the BMC web UI, or POST to /redfish/v1/Managers/1/LicenseManager/QueryLicense. There is no Redfish LicenseService on Supermicro, so licence state cannot be read with a plain GET.

Warning

Keys are bound to the BMC MAC address and are neither transferable nor refundable. Put the part number in the purchase order rather than deciding later.

Getting to the configuration utility

Power on and press DEL repeatedly at the Supermicro splash screen to enter the BIOS setup utility. F11 gives the one-time boot menu instead, which is not what you want here.

• Onboard SATA: Advanced > PCH SATA Configuration, set SATA Mode to AHCI. Most X11 and X12 boards also list a separate PCH sSATA Configuration entry driving a second set of ports, and both must be set. Setting only the first is the most common miss on Supermicro.

• Direct NVMe: Advanced > Chipset Configuration > North Bridge > IIO Configuration > Intel VMD Technology, leave Disabled. Never create a VROC volume. AMD H12 and H13 have no VMD or VROC.

• Broadcom RAID card: Advanced > the Broadcom or AVAGO configuration utility.

The Advanced tab on an X11 board, Aptio Setup Utility 2.20.1271. Two entries matter here and they are separate menus: PCH SATA Configuration and PCH sSATA Configuration. Each drives its own set of ports and each has its own mode setting, so visiting only the first leaves half the ports unconfigured. Note also what is absent: there is no Broadcom or AVAGO utility in this list, because no such card is installed in this chassis. On a board with onboard SATA only, the list ends at Driver Health and these two menus are the entire job.

Set both SATA controllers to AHCI

Advanced > PCH SATA Configuration. The field that decides pass-through is Configure SATA as, which reads [AHCI] here. Required state: AHCI. Read the port list below it as well: every port on this controller reports Not Installed, so no drive is attached to it on this machine. That is the reason the second controller cannot be skipped.

Advanced > PCH sSATA Configuration, with Configure sSATA as opened to show the full choice. The control offers exactly two values, AHCI and RAID. Required state: AHCI. Selecting RAID puts this controller in the data path and ZFS no longer receives raw disks, which is the same outcome the Broadcom section below exists to prevent. Note that sSATA Port 0 carries the only drive in this machine, a KINGSTON SEDC5 of 480 GB, while the PCH SATA screen above showed no drives at all. An operator who set only the first controller would have configured the empty one and left the populated one on whatever it shipped with.

Tip

Both SATA modes can be read, and set, over the network with no reboot. The BMC exposes them as BIOS attributes:

GET /redfish/v1/Systems/1/Bios

Read ConfigureSATAas (the PCH SATA controller) and ConfiguresSATAas (the PCH sSATA controller). Each is an enumeration with exactly two values, AHCI and RAID. Both must read AHCI. RAID on either one means that controller builds a RAID layer and ZFS does not receive raw disks, which is the same failure the Broadcom section below exists to prevent, on different silicon.

Confirm the allowed values for a given board from /redfish/v1/Registries/BiosAttributeRegistry.v1_0_0. Ignore the HelpText field there: Supermicro’s registry misattributes it, and for these two attributes it shows the help belonging to SATADeviceType.

While reading the same response, check every Intel VMD for Volume Management Device attribute is Disable.

Note

The Broadcom entry appears in the Advanced menu only when a Broadcom card is physically installed. On a node with onboard SATA only, the Advanced list ends at Driver Health with no controller utility, and that is the expected, correct state: there is no RAID layer to disable. Confirm it rather than hunting for a hidden menu, either from the BMC at GET /redfish/v1/Chassis/1/PCIeDevices (a storage card appears as its own member alongside the NICs) or from a running OS with lspci | grep -iE 'lsi|broadcom|sas'. Where neither reports a card, the only BIOS work is the two SATA mode settings above.

Supermicro does not ship its own RAID firmware. The card in the chassis is a Broadcom part with Supermicro’s OEM sticker, and the passthrough procedure depends on which Broadcom silicon is present.

First check: what is actually in the box.

From a running Linux host:

lspci | grep -iE 'lsi|broadcom|sas'

Three outcomes. Each maps to a different section of this walk-through:

What lspci reports	What to do
Only onboard Intel C6xx SATA / sSATA	Nothing to configure. The onboard chipset is already in AHCI passthrough for ZFS.
Broadcom SAS3008 (onboard, or an -L8e add-on card)	True IT-mode firmware is available for this silicon. That is the cleanest state for ZFS.
Broadcom SAS3108 (typical on -H8iR add-on cards)	No IT-mode firmware exists for this silicon. JBOD on a per-drive basis is the best available, and cache settings become load-bearing.

Supermicro card naming. The suffix on Supermicro’s part number tells you what class the card is:

• -L8e = IT / HBA class. This is the IT-mode SKU.

• -L8i = IR (integrated RAID). Broadcom SAS3008 underneath, can be flashed to IT firmware.

• -H8iR = SAS3108 hardware RAID. JBOD only. Cannot be flashed to IT.

Confirm the suffix on the purchase order. -H8iR and -L8e are the same form factor and the difference is not visible once the card is installed.

Path A: SAS3008 (``-L8e`` or ``-L8i`` flashed to IT).

IT-mode firmware removes the RAID stack entirely. The card behaves like a plain HBA: no cache in the data path, no controller metadata on the drives, full SMART, and the drives stay portable to any other HBA. This is the cleanest configuration for ZFS.

Confirm from Linux:

lsblk -d -o NAME,SIZE,MODEL,SERIAL

Every row should carry the drive’s real model and serial. Nothing else is needed on the controller.

Path B: SAS3108 (JBOD, no IT mode available).

Two settings must both be applied - enabling JBOD Mode on the controller does not by itself convert any drive.

1. Enable JBOD Mode on the controller (Broadcom UEFI HII: Main Menu > Controller Management > Advanced Controller Properties > JBOD Mode > Enabled), then Apply Changes and reboot.

2. Convert each drive to JBOD state (Broadcom UEFI HII: Main Menu > Drive Management > [drive] > Operations > Make JBOD).

From StorCLI on a running host:

storcli /c0 show
storcli /c0 set jbod=on
storcli /c0/eall/sall set jbod
storcli /c0/eall/sall show

Warning

On some newer Broadcom firmware the toggle is spelled set autoconfig=jbod instead of set jbod=on. If storcli returns a syntax error, try the other form.

Warning

If a drive is not in Unconfigured Good state, StorCLI will refuse the JBOD conversion. Set the drive state first with storcli /c0/e<enc>/s<slot> set good then retry.

Cache settings on Path B. With the SAS3108 still in the data path, controller cache policy affects ZFS’s write-flush guarantees. Set virtual-disk cache to write-through and read-ahead off:

storcli /c0/vall set wrcache=WT
storcli /c0/vall set rdcache=NoRA
storcli /c0/vall show all | grep -iE 'cache|policy'

Verify from the OS.

On either path:

lsblk -d -o NAME,SIZE,MODEL,SERIAL

If any row’s model shows MegaRAID or LSI ROMB, JBOD has not fully taken effect. Recheck steps 1 and 2 for Path B or reflash to IT firmware for Path A.

Note

Supermicro’s BMC (IPMI 2.0 web UI) does not expose a RAID configuration interface. Newer BMC firmware shows storage inventory only, not configuration. Remote-only workflows use StorCLI over SSH into a running OS, or IPMI serial-over-LAN into the pre-boot BIOS.

Verify from Linux

lsblk -d -o NAME,SIZE,MODEL,SERIAL,WWN,TRAN
lsblk -dD -o NAME,DISC-GRAN,DISC-MAX
storcli /c0/vall show     # must report no virtual drives

The driver stays megaraid_sas even in JBOD, so the driver name proves nothing. Judge on the model string and the discard value.

Dell

BMC licence required

iDRAC9 Enterprise or Datacenter

14G, 15G and 16G need iDRAC9 Enterprise. 17G needs iDRAC10 Enterprise, and iDRAC10 Core is not sufficient. 13G needs iDRAC8 Enterprise. iDRAC9 Basic and Express are not sufficient.

Check with GET /redfish/v1/LicenseService/Licenses, or the iDRAC UI under Licenses.

Warning

An under-licensed Dell looks completely healthy. Basic and Express still carry Redfish, power control, boot control, inventory and firmware update, so the node registers and inventories perfectly and fails only at the ISO mount. Licences are locked to a server generation and cost more after point of sale.

Verified on PowerEdge R640 with PERC H740P Mini.

Two screens, told apart by the breadcrumb at the top:

• Screen 1 - breadcrumb reads Dashboard View > Main Menu > Manage Controller Mode. Shows Controller Mode and Auto-Configure Behavior.

• Screen 2 - breadcrumb reads Dashboard View > Main Menu > Configuration Management. Shows a list: Create Virtual Disk / Convert to RAID Capable / Clear Configuration / Manage Foreign Configuration.

There is no third screen showing Convert to Non-RAID Disk. When every disk is already Non-RAID, that menu item does not exist. This is expected.

How to get there.

1. Power on the server and press F2 at the Dell splash screen. This enters System Setup.

2. Select Device Settings.

3. Select Integrated RAID Controller 1: Dell PERC H740P Mini Configuration Utility.

4. Select Main Menu.

Step 2, System Setup > Device Settings on a PowerEdge R640. The storage controller appears as Integrated RAID Controller 1: Dell PERC H740P Mini Configuration Utility, above the four integrated NIC ports. Select that entry. Only devices conforming to UEFI HII appear in this menu, so a controller missing from this list is either absent or the server is in legacy boot mode.

Step 4, the controller utility’s Dashboard View > Server Profile. Four entries appear under CONTROLLER COMPONENTS: Controller Management, Hardware Components, Physical Disk Management and Virtual Disk Management. Mode is read and set under Controller Management; the other three report state only.

Step 1: Check the mode on the Manage Controller Mode screen.

Navigate: Main Menu > Controller Management > Manage Controller Mode. The breadcrumb should read Dashboard View > Main Menu > Manage Controller Mode.

Controller Management, which carries both the firmware level and the fastest state check on this controller. Read the three disk counts: Physical Disk Count 5, Non-RAID Disk Count 4, Virtual Disk Count 0. Required state: Virtual Disk Count is 0 and every data drive is counted as Non-RAID. A non-zero virtual disk count means an array still exists and must be cleared before the controller will leave RAID mode. Advanced Controller Management, at the bottom of this page, is where the mode itself is changed.

Read the two fields:

• Controller Mode should read Enhanced HBA.

• Auto-Configure Behavior should read Non-RAID Disk.

If both read exactly that, the machine is already correct for ZFS. Change nothing and exit.

Warning

Do not click Switch to RAID Controller Mode. That link is on this screen and it does the opposite of what you want; it turns RAID back on. Its presence means you are already in the right mode.

The Advanced page carries a second, more destructive control: Force Switch to RAID Controller Mode, which switches the controller back to RAID even when a configuration is present. The on-screen warning is explicit that Non-RAID Disks will be lost and that all virtual disks become foreign. Dell states it should only be used as part of a controller replacement. Neither this control nor the plain Switch to RAID Controller Mode link is ever used when preparing a node for Karios.

Step 2: Only if Controller Mode reads RAID.

On the same screen the link will read Switch to Enhanced HBA Mode instead.

1. Select Switch to Enhanced HBA Mode and confirm.

2. Set Auto-Configure Behavior to Non-RAID Disk.

3. Reboot.

If the switch is refused, virtual disks still exist. Delete them first: Main Menu > Configuration Management > Clear Configuration.

Warning

Clear Configuration destroys every virtual disk and all data on them. Back up first. Then repeat Step 2.

Step 3: Only if some disks are still not Non-RAID.

Navigate: Main Menu > Physical Disk Management. Select a disk, then Operations > Convert to Non-RAID disk > Go > confirm. Repeat per disk.

In Enhanced HBA mode with Auto-Configure Behavior set to Non-RAID Disk, this is usually unnecessary. New disks come up Non-RAID on their own.

About Screen 2: Configuration Management.

This screen is not where you check or change the mode. It looks nearly identical whether the controller is in RAID mode or Enhanced HBA. Always decide from Screen 1 (Manage Controller Mode), not from here.

The list on Screen 2 shows Create Virtual Disk, Convert to RAID Capable, Clear Configuration, Manage Foreign Configuration. The absence of Convert to Non-RAID Disk is intentional: Dell hides conversion options that have nothing to act on. When every disk is already Non-RAID, only the reverse direction is offered.

What you see on Screen 2	What it means
Convert to RAID Capable present, Convert to Non-RAID Disk absent	All disks are already Non-RAID. Nothing to do.
Convert to Non-RAID Disk present	Disks still need converting. Use Step 3.
Neither present	Disks belong to a virtual disk. Delete it first with Clear Configuration.

Step 4: Verify from the OS.

After the installer completes and the node reboots, run:

lsblk -d -o NAME,SIZE,MODEL,SERIAL

You should see one row per physical disk, each with its real model and serial number. If you see a single large device, or a device named PERC H740P Front or LOGICAL VOLUME, the controller is still presenting a virtual disk. Go back to Step 1.

Requirements.

• PERC firmware 50.5.0-1750 or later for Enhanced HBA on the H740P. Check under Controller Management > View Controller Information. Update from iDRAC or Lifecycle Controller if older.

• Enhanced HBA allows virtual disks alongside Non-RAID disks, so a small RAID-1 boot pair plus passthrough data drives is supported and is usually the right layout.

• The PERC H755 has no Enhanced HBA mode at all. Per-disk Non-RAID is the best available there.

Other PERC generations

The walkthrough above covers the PERC 10 H740P. The rest of the family differs:

HBA355i, HBA330, HBA465i	already pass-through	nothing to set
PERC 10: H740P, H745, H745P MX	Enhanced HBA (eHBA)	yes, as above
PERC 10: H345	Non-RAID per disk	no eHBA on this card
PERC 10: H840	none exists	cannot run Karios
PERC 11: H755, H355, H350	Non-RAID per disk	no controller mode
PERC 12/13: H965i, H365i, H975	Non-RAID per disk	no controller mode
PERC 9: H330, H730, H730P	HBA mode	yes
PERC 7/8: H310, H710, H710P	none	refuse at sizing

On PERC 11, 12 and 13 there is no controller mode at all. Use Configuration Management > Convert to Non-RAID Disk, per drive, and set Auto-Configure Behavior so newly inserted drives arrive Non-RAID. Because no single field proves the state, every drive has to be checked.

Warning

PERC 9, 10 and 11 use perccli64. PERC 12 and 13 use perccli2. A script written for one will not run on the other.

Tip

Put an HBA355i on a new Dell quote and this whole tab becomes unnecessary. Dell states the HBA355i family does not support RAID volumes at all.

HPE

BMC licence required

iLO Advanced

The gated feature is named exactly Scripted or URL-Based Virtual Media, and it is iLO Advanced only. iLO Standard does not have it.

Check with GET /redfish/v1/Managers/1 and read Oem.Hpe.License.LicenseString.

Warning

Two ways to buy the wrong thing. iLO Essentials ships bundled with the iLO network enablement module P13788-B21, exactly the option an integrator adds to get a dedicated management port, and Essentials carries console virtual media but not the scripted media Karios uses. Separately, an iLO Advanced licence is necessary but not sufficient: Virtual Media can still be switched off under Access Settings.

Verified on an HPE ProLiant Gen10 with a Smart Array P408i-a SR Gen10, controller firmware 8.23.

Getting to the configuration utility

Two routes at power-on. Either one reaches the setting; you do not need both.

• ``F10`` > Intelligent Provisioning > Perform Maintenance > HPE Smart Storage Administrator. The graphical route, and the one used throughout this tab.

• ``F9`` > System Utilities > System Configuration > the controller. The UEFI route, useful when Intelligent Provisioning has been disabled.

The whole procedure in one line:

POST -> F10 -> Intelligent Provisioning -> HPE SSA -> Clear Configuration -> Controller Mode = HBA

Tip

The controller mode can be read over the network, with no reboot. This is the fastest way to survey a fleet before anyone visits a console:

GET /redfish/v1/Systems/1/SmartStorage/ArrayControllers/

Follow each member and read Model, FirmwareVersion.Current.VersionString and CurrentOperatingMode. CurrentOperatingMode returns RAID, Mixed or HBA directly, which answers the question this whole tab exists to answer. Only HBA is correct for Karios.

Step 0: identify the controller tier

Smart Array SR Gen10 splits into three tiers, and the tier decides how much work there is to do. Read the controller name on the SSA device list, or from iLO under System Information > Firmware > Storage Controllers.

Tier	Example	What is required
P-series	P408i-a, P816i-a	Has an explicit Controller Mode with values RAID and HBA. Steps 1 to 4 below apply.
E-series	E208i-a	HBA-class by design, with no write cache and no mode to change. Drives reach the OS directly. Nothing to configure.
S-series	S100i SR Gen10	Onboard software SATA. It has no true pass-through path on Linux. Disable it in system BIOS and set the onboard SATA ports to AHCI instead.

All three tiers use the smartpqi kernel driver on Gen10 and later. Gen9 and earlier use hpsa. Neither driver name tells you which mode the controller is in, so the driver is not a check.

Step 1: clear the existing configuration

A controller cannot leave RAID mode while logical drives exist on it.

1. In SSA, select HPE Smart Array P408i-a SR Gen10 from the device list on the left.

2. In the centre action column, select Configure.

3. Select Clear Configuration, confirm, then select Finish.

Warning

Clear Configuration permanently deletes every logical drive on the controller, along with its metadata and the data it holds. Anything that must survive has to be backed up or physically removed before this step.

Step 2: confirm no logical drive remains

Configure > Create Array on a P408i-a SR Gen10, used here only to read the drive inventory. This screen lists physical drives that do not belong to a logical drive, so every bay appearing here is unassigned. Six 400 GB SAS SSDs are present, four on Internal Drive Cage at Port 1I : Box 1 and two on Port 2I : Box 1. Required state: every drive physically installed in the chassis appears in this list. A populated bay that is missing here still belongs to a logical drive, so repeat Step 1. Select Cancel when the count matches. Creating an array here would undo Step 1.

Step 3: set Controller Mode to HBA

1. Return to the controller configuration pane, select Configure, then Modify Controller Settings.

2. Change Controller Mode from RAID to HBA.

3. Select Apply. SSA prompts for a restart.

4. Exit SSA and reboot the host.

On the UEFI route the same setting is at F9 > System Configuration > the controller > Configure Controller Settings > Modify Controller Settings > Controller Mode > HBA, saved with F12.

Mixed Mode is not HBA Mode

A P-series controller can hold logical drives and unassigned drives at the same time, a global state HPE calls Mixed Mode. The unassigned drives are visible to the operating system and look passed through, but the controller remains in the data path for the array it still owns. Karios requires the controller out of the data path entirely, so the global Controller Mode must read HBA.

HPE Smart Storage Administrator 3.40.3.0, controller main page for a Smart Array P408i-a SR Gen10. Everything on this page looks correct for pass-through: Logical Devices reads 0 arrays, 0 logical drives, Unassigned Drives reads 6 unassigned drives, and the Controller Configuration Summary confirms 0 Data Array(s) and 6 Unassigned Drive(s). It is not sufficient. Under Port Settings, Port 1I and Port 2I both report Mixed Mode, which is the global controller state, and the drives are unassigned while the controller itself is still in RAID. Required state: Controller Mode = HBA, a value this page never displays. Confirm it on Configure > Modify Controller Settings, and if it reads RAID there, apply Step 3.

Step 4: verify from the operating system

Once the host has booted, query the block layer:

lsblk -d -o NAME,SIZE,MODEL,SERIAL

Pass-through succeeded. Each row carries a real HPE model string and a distinct drive serial:

NAME   SIZE  MODEL           SERIAL
sda    1.8T  VK001920GWTTE   2XG4Y7R1
sdb    1.8T  VK001920GWTTE   2XG4Y7R8

Pass-through failed. The controller is still presenting a logical volume, and the individual drives are hidden behind it:

NAME   SIZE  MODEL               SERIAL
sda    3.6T  HPE LOGICAL VOLUME  600508b1001c...

A model string of HPE LOGICAL VOLUME means Step 3 has not taken effect. Repeat it and reboot. The same string appears on Gen9 under hpsa.

Firmware

Target controller firmware v7.43 or v8.00 and later for production. Check it in iLO 5 under System Information > Firmware > Storage Controllers > HPE Smart Array P408i-a SR Gen10, and update from the Service Pack for ProLiant before installing Karios rather than after.

Adding a drive to a running node

In HBA mode the controller presents a newly inserted drive to the backplane immediately, but the Linux kernel does not always rescan on its own. Rather than rebooting a node that is carrying Ceph or ZFS data, force the rescan:

echo "1" > /sys/class/scsi_host/host0/rescan

Replace host0 with the SCSI host that maps to the controller. List the candidates with ls /sys/class/scsi_host/.

Gen11 and iLO 6

The Intelligent Provisioning workflow above is the same on Gen11 under iLO 6. Gen11 adds the SR932i-p and SR416i-a Gen10 Plus and Gen11 controllers, which use a refreshed interface layout, so the labels move but the sequence does not.

Gen11 also ships MR controllers, covered next, which are a different product family entirely.

Gen11 MR series, which does not pass drives through by default

The section above covers the Gen10 SR family. Gen11 also ships MR controllers, which are Broadcom MegaRAID and behave nothing alike.

HPE states it plainly: “For HPE MR200 controllers, by default it is set to JBOD”, but “For HPE MR400 controllers, by default it is set to None”, and “If Auto-Config is set to None, the new drive status stays in Unconfigured Good”.

A factory MR416i does not pass drives through at all, and there is no HBA mode to click. Set it explicitly:

1. F9 > System Configuration > HPE MR416i Gen11 > Configuration Management > Clear Configuration.

2. Controller Management > Advanced Controller Management > Manage Personality Mode, set Auto-Configure Behavior to JBOD.

3. Convert each drive to JBOD.

Danger

This is the card that failed in the field: HBA mode did not take, RAID virtual disks ended up under ZFS, IO stalled and Ceph froze, while an identical node with true pass-through stayed healthy. Prove it with lsblk -dD.

Warning

Workload profiles LowLatency, HighPerformanceCompute and GraphicProcessing force ProcVirtualization and IntelProcVtd to Disabled and read-only, disabling the virtualization Karios requires. Use GeneralPowerEfficientCompute.

Cisco

BMC licence required

None

Cisco IMC is included with every UCS C-Series rack server at no additional charge. No licence tier, feature key or add-on gates vKVM, virtual media, Redfish or IPMI.

Nothing to check. Confirm instead that the server is in standalone Cisco IMC mode.

Warning

The mode matters instead. A server in Intersight Managed Mode, or attached to Fabric Interconnects, cannot be driven by Karios. That is decided at order time or at first power-on.

Getting to the configuration utility

Cisco is the only manufacturer here where this can be done entirely out of band, with no reboot and no console trip. Log in to the CIMC web interface and go to Storage > Controllers.

Note

On M7 the menu moved. It is Apps > Storage > Controllers.

If you do want the BIOS instead, press F2 at POST for setup, or F6 for the one-time boot menu.

Verified on UCSC-C240-M5 with Cisco 12G Modular RAID Controller (4 GB cache, max 26 drives), firmware 07.23.03.00 / package 51.23.0-5009.

Important

This procedure is different from Dell’s in three ways that matter. Read them before starting; applying the Dell steps here will fail.

Dell PERC H740P	Cisco 12G Modular
Controller Mode is settable (RAID vs Enhanced HBA)	Personality Mode is read-only, permanently RAID
One screen holds mode and behaviour together	Mode display and behaviour setting are on different screens
Reached by F2 > Device Settings	Reached by F2 > Advanced tab (UEFI) or Ctrl+R (legacy boot)

How to get there.

1. Power on and press F2 to enter the Aptio Setup Utility.

2. Press right-arrow to the Advanced tab.

3. Scroll to the bottom entry: AVAGO <Cisco 12G Modular Raid Controller with 4GB cache (max 26 drives)> Configuration Utility - 07.23.03.00.

4. Enter, then Dashboard View > Main Menu.

Note

If that entry is absent, the server is in legacy boot mode. Reboot and press Ctrl+R at the controller banner instead; same controller, older key-driven interface.

Step 1: Read the current state.

Navigate: Main Menu > Controller Management. Read three fields:

• Controller Status (typically Needs Attention).

• Personality Mode (RAID, read-only).

• Select Boot Device.

Personality Mode is read-only. Pressing Enter on it does nothing. The help pane confirms: “Displays the current personality mode”. Compare Dell’s Manage Controller Mode screen, where the help reads “Selects behavior mode”; that verb is a reliable test for whether any field in these utilities is editable.

Check Select Boot Device before going further. If it names a virtual drive, an OS lives on it and the next step destroys it.

Step 2: Clear the configuration.

Navigate: Main Menu > Configuration Management > Clear Configuration. A warning appears:

“Clear Configuration will delete all of the Virtual Drives, Hot Spare Drives, Pinned Cache and JBODs (if applicable) attached to this controller.”

It uses a two-step guard:

1. Set Confirm from [Disabled] to Enabled.

2. Then select Yes.

Yes has no effect while Confirm reads Disabled, and the utility gives no feedback to say why. Set Confirm first.

Step 3: Set the auto-configure behaviour.

Navigate: Main Menu > Controller Management > Advanced Controller Management > Manage Personality Mode.

That page holds three items:

• Personality Mode [RAID] - read-only, ignore.

• Auto-Configure Behavior [None] - this is the one.

• Apply Changes.

Set Auto-Configure Behavior to JBOD, then select Apply Changes.

Warning

If the options include Auto-configure RAID 0, do not pick it. That is the single-disk-RAID-0 trap: the controller stays in the data path, writes RAID metadata to each drive, and hides SMART.

Warning

Set Factory Defaults sits directly above Manage Personality Mode in the Advanced Controller Management list. Selecting it resets the controller and discards the work done in the previous steps.

Step 4: Convert any remaining drives.

Navigate: Main Menu > Drive Management. Select a drive, then Operations > Make JBOD > Go.

Auto-Configure only acts on unconfigured drives, so anything left over from before Step 2 may need this. Every data drive should end at state JBOD.

Drive state	Meaning
JBOD	Passed through to the OS.
Unconfigured Good	Released from RAID but not passed through. Still needs Make JBOD.
Online	Still in a virtual drive.
Foreign	Carries another controller’s config. Clear foreign config first.

Step 5: Save.

Esc back to the BIOS, press F10 and choose Save & Reset System.

Verify.

After reboot, run:

lsblk -d -o NAME,SIZE,MODEL,SERIAL

Count the drives. Real model and serial per row. If any drive is missing, find that now rather than mid-install.

The out-of-band route, step by step

CIMC > Storage > the MRAID controller > Controller Info, reached without rebooting the server. Product PID identifies the exact part, here UCSC-RAID-M5HD, which is the string to match against a Cisco quote or against the generation table further down this tab. Under Settings on the right, JBOD reports whether the controller will export drives at all; it reads true on this machine. Required state: JBOD: true. Where it reads false, apply Actions > Enable JBOD before converting any drive, since the per-drive action is unavailable until the controller permits it.

Physical Drive Info on the same controller, listing per-drive state. The controller already reports JBOD: true, yet the State column is mixed: some drives read JBOD and others read unconfigured good. Only the drives in JBOD state are exported to the operating system; unconfigured good means the drive is free of any virtual drive but is not yet passed through. Required state: JBOD on every drive. Select the remaining drives and apply Enable JBOD from the button row above the table.

1. Select the LSI MegaRAID controller, then the Controller Info tab.

2. Actions > Enable JBOD, and confirm.

3. If the drives carry another controller’s configuration, Actions > Clear Foreign Config first.

4. Go to Physical Drive Info. For each drive, open the three-dot menu, choose Set State as Unconfigured Good if it is Online, then Set State as JBOD.

Other Cisco generations

UCSC-RAID-M5, UCSC-RAID-M5HD	RAID	JBOD, controller then per drive
UCSC-SAS-M5, UCSC-SAS-M5HD	pure HBA	nothing to set
UCSC-RAID-M6T, UCSC-RAID-M6HD	RAID	JBOD
UCSC-SAS-M6T, UCSC-SAS-M6HD	pure HBA	nothing to set
UCSC-HBA-M1L16 (M7)	pure HBA	nothing to set
UCS-M2-HWRAID (boot)	boot carrier	JBOD, also the factory default

Tip

Specify a UCSC-SAS- part rather than a UCSC-RAID- part in the same chassis slot and this whole procedure disappears.

Warning

Firmware images are model-exact. A C240 M5L is not a C240 M5, and a C220 HUU will not upgrade a C240.

Lenovo

BMC licence required

XCC Enterprise, XCC2 Platinum, or XCC3 Premier

The tier depends on the controller generation. XCC1 (ThinkSystem Gen1 to Gen3) needs Enterprise. XCC2 (V3) needs Platinum, FoD part 7S0X000KWW. XCC3 (V4) needs Premier, 7S0X000XWW.

Check with GET /redfish/v1/Managers/1/Oem/Lenovo/FoD/Keys, the XCC UI under BMC Configuration > License, or keycfg over SSH.

Warning

Standard is never sufficient, and on XCC1 Advanced is not sufficient either. The XCC1 upgrade route also splits between a direct upgrade and a two-step pair depending on machine type, so the wrong route produces an unbuildable order.

Getting to the configuration utility

Power on and press F1 at the Lenovo splash screen to enter UEFI Setup. On newer machines this opens XClarity Provisioning Manager, from which you choose UEFI Setup. Then go to System Settings > Storage and select the adapter.

UEFI Setup on a ThinkSystem SR630, System Settings > Storage, which lists every storage adapter the machine can see. This node carries two: an AVAGO MegaRAID SAS 9364-8i in slot 4 and a ThinkSystem M.2 with Mirroring Enablement Kit in slot 5. Each is configured separately, and the two belong to different controller families with different mode names and different tools, so identify both from this screen before opening either utility.

Work out which family the card belongs to

Lenovo is the only manufacturer in this section that ships two controller families with different mode names and different configuration tools. The setting that works on one family does not exist on the other, so identify the family from the adapter name before opening any utility.

Family	Parts	Driver	Mode name	Tool
Broadcom MegaRAID	530, 540, 545, 730, 930, 940, and the AVAGO 9364-8i	megaraid_sas	JBOD	StorCLI
Microchip Adaptec	5350, 9350	smartpqi	HBA mode	ARCCONF

Parts with no RAID mode to disable, and therefore nothing to configure: 430-8i, 430-16i, 440-8i, 440-16i, 4350-8i. The 440 series is the part to quote on a new order.

Broadcom MegaRAID cards, set JBOD

RAID Setup > Select RAID Adapter, which lists every adapter with its current virtual-disk count. The AVAGO MegaRAID SAS 9364-8i in slot 4 reports 1 virtual disk created, and the configuration table below it resolves that to Array0, a RAID 1 of 237 GB. The M.2 Mirroring Kit in slot 5 reports No virtual disks created. Required state: No virtual disks created on every adapter, because a controller will not release drives that belong to an array. Where an adapter reports one or more virtual disks, delete them through Configuration Management > Clear Configuration before continuing.

Manage Disk Drives on the same adapter, showing per-drive state. Two drives read Online, which means they are members of the Array0 RAID 1 above, and two read UGood (Unconfigured Good), which means they are free of any array but not yet exported. Required state: JBOD on every drive Karios will use. Enabling JBOD Mode on the controller changes the controller’s default for drives inserted afterwards; it does not convert drives already installed, so each one is converted individually in the steps below.

1. Controller Management > Advanced Controller Properties > JBOD Mode > Enabled.

2. Configuration Management > Clear Configuration, which destroys all virtual disks.

3. Drive Management, each drive, Unconfigured Good then Make JBOD.

storcli64 /c0/v delete
storcli64 /c0 set jbod=on
storcli64 /c0/eall/sall set good
storcli64 /c0/eall/sall set jbod

Microchip Adaptec cards, set HBA mode

Controller Management > Modify Controller Settings > Configure Controller Mode set to HBA Mode, after deleting any logical drives. The Linux tool here is arcconf, not storcli.

Warning

Mixed Mode is not HBA mode. On the 9350 series Mixed Mode allows RAID and HBA devices simultaneously, so unassigned drives look passed through while the controller stays in the data path. Same trap as the HPE Smart Array P-series.

Verify from Linux

lsblk -d -o NAME,SIZE,MODEL,SERIAL,WWN
lsblk -dD -o NAME,DISC-MAX

No drive may show a naa.600605b WWN prefix, and none may read MegaRAID, ServeRAID or LOGICAL VOLUME.

Boot device

Danger

The Karios installer’s boot-mirror exemption is a literal match on BOSS and NS204 only. Lenovo M.2 RAID parts get no exemption. Set the M.2 adapter to JBOD, not RAID 1, and let ZFS mirror the pair. The M.2 with Mirroring Enablement Kit ships defaulted to RAID 1, so an untouched kit arrives in the wrong state.

Tip

The cleanest Lenovo node for Karios is an SR630 V3 or V4 with an M.2 2-Bay kit in JBOD and no SAS or RAID controller at all.

7. Before You Begin

Confirm the following before starting the installation:

• Hardware meets baseline requirements

• Installation media is ready

• The Karios platform image (.qcow2, approximately 5 GB) is downloaded onto a second machine that can reach the node over the management network

• That second machine has a browser available - the install pauses partway through and waits for the image to be delivered from it

• Management network is reachable

• You have console access to target hardware

8. Platform User Flow

This is a suggested flow to help you get started. You are free to follow your own order based on your deployment needs. The actual flow will vary depending on your objective.

After installation, review the Control Center Dashboard and Infrastructure to confirm the environment is healthy, then go to User Management to create users and assign roles before proceeding further.

If you have purchased a license, go to User Management → License to review the license overview and complete the activation flow before creating users. See 10. License.

For example, if your immediate objective is to launch a VM, you can go directly to Compute – Virtual Machines after reviewing Infrastructure. The flow is entirely driven by your objective.

ISO Install (9-step wizard)
→ Control Center Dashboard
→ K-Store  (Platform updates → Plugins)
→ Infrastructure  (Zone → Pods → Clusters → Hosts)
→ User Management  (License activation → Create users and assign roles)
→ K-Forge  (Add Host → Reveal → Configure)
→ Network
→ Storage
→ Compute – Virtual Machines  (Create VM → Manage VM)
→ Nodes  (Review / Add Hosts)
→ Compute Policies
→ Observability
→ K-Shield  (Security)
→ K-Trace  (Security Operations)
→ Kubernetes
→ Karios DFS
→ Release Management  |  Settings

7.1. Section Links

1. ISO Install

2. Infrastructure (Zone → Pods → Clusters → Hosts)

3. User Management (Create users and assign roles)

4. K-Forge (Add Host → Reveal → Configure)

5. K-Nets

6. Storage

7. Compute – Virtual Machines (Create VM → Manage VM)

8. Nodes (Review / Add Hosts)

9. Compute Policies

10. Observability

11. K-Shield (Security)

12. K-Trace (Security Operations)

13. Kubernetes

14. Karios DFS

15. Release Management | Settings

9. Karios DFS Entry Point

Use the Karios DFS section when you begin storage-cluster operations.

Recommended DFS entry order:

1. Karios DFS Overview (roles, dependency chain, dashboard health check)

2. Storage (Pools, File System, NFS Shares)

3. Block Storage (Block Images; iSCSI and Mirroring are in beta)

4. Object Storage (Users, Gateway, Buckets)

5. Infrastructure, Monitoring, and Advanced as needed

Note

In Karios DFS terminology, use NFS Shares for the Storage NFS page and Manager Modules for Advanced > Extensions.

10. Getting Help & Support

If you are blocked, use one of these paths:

• Email: support@karios.com

• Portal: Karios Support

Include in every request:

• Exact error text

• Timestamp

• Doc section and step

• Current impact

Note

Full support workflow and ticketing details are documented in Karios Support.

11. Success Checkpoint

After completing Getting Started, you should be able to:

• Explain the first-time onboarding path

• Identify minimum hardware and network prerequisites

• Choose the correct support escalation path and required evidence

• Proceed directly to ISO installation with clear expectations