Re: ssd6202a RAID card and Clonezilla?

On 11/1/25 2:16 AM, Paul wrote:
> On Sat, 11/1/2025 4:14 AM, T wrote:
>> Hi All,
>>
>> Anyone know if this RAID card will work with clonezilla.
>> I know most will not.
>>
>> https://www.highpoint-tech.com/product-page/ssd6202a
>>
>> Their web site states:
>>
>>      HighPoint Industrial and Virtualization solutions feature integrated host, hardware-level RAID 0, 1 and JBOD capability. RAID arrays hosted by SSD6200 series controllers are fully transparent and will be automatically recognized by all current VM and OS platforms as available single-drives. No drivers, special procedures or software application is required.
>>
>> Yours in Confusion,
>> -T
> 
>   https://www.techpowerup.com/327842/highpoint-unveils-new-ssd6202a-low-profile-aic-nvme-boot-drive-solution
> 
>    "Same Marvell 88NR41 controller as in Western Digital´s WD AN1500."
> 
>    # Google gives me the wrong URL, but you get the idea.
> 
>    https://www.marvell.com/products/system-solutions/nvme-accelerators.html
> 
>       "By extending Marvell’s SATA/SAS legacy controllers' feature set into the NVMe space"    # Not according to CoPilot, oh well
>                                                                                                # A mis-leading blurb.
> 
> CoPilot Question
> ----------------
> 
> How does a marvell 88NR41 NVMe RAID devices present itself
> to the OS, when the hardware is being discovered,
> and the card is operated in RAID1 mode or JBOD mode ?
> 
> CoPilot answer
> --------------
> 
> **In RAID1 mode, the Marvell 88NR41 NVMe RAID device presents a single virtual disk to the OS;
>    in JBOD mode, each physical NVMe SSD is exposed individually.**
> 
> Here’s how the Marvell 88NR41 (often referenced as 88NR2241 in product literature)
> behaves during hardware discovery and OS enumeration:
> 
> ### Device Presentation to the OS
> 
> #### **RAID1 Mode**
> - **Virtualization Layer:** The controller aggregates two physical NVMe SSDs into a single *virtual disk*.
> - **OS View:** The operating system sees **one logical NVMe device**, representing the mirrored pair.
> - **Driver Role:** Marvell’s NativeRAID firmware and driver stack handle the RAID logic internally, abstracting redundancy from the OS.
> - **Boot Support:** This virtual disk can be used as a boot device, especially in server environments where redundancy is critical.
> 
> #### **JBOD Mode**
> - **Direct Mapping:** Each NVMe SSD is passed through directly to the OS without abstraction.
> - **OS View:** The OS enumerates **each physical NVMe drive independently**, as if they were connected directly via PCIe.
> - **Use Case:** Ideal for software RAID setups or when individual drive access is needed for performance or flexibility.
> 
> ### Discovery Process
> 
> - **PCIe Enumeration:** The card appears as a PCIe device with multiple endpoints depending on configuration.
> - **NVMe Controller Exposure:** In RAID1, the virtual disk is exposed via a virtual NVMe controller. In JBOD, each SSD’s native controller is visible.
> - **CLI Tools:** Marvell provides command-line utilities to inspect physical and virtual disk mappings, useful for verifying RAID status and drive health.
> 
> ### Additional Notes
> 
> - **DRAM-less Architecture:** The 88NR41 uses a DRAM-less design, relying on host memory and low-latency NVMe transactions.
> - **RAID Modes Supported:** RAID0, RAID1, RAID10, and JBOD are all supported, allowing flexible deployment strategies.
> - **Target Use Cases:** Boot OS replication, hyperconverged infrastructure, and high-availability storage systems.
> 
> Would you like a breakdown of how the virtual disk metadata is structured or
> how the firmware handles failover in RAID1? I can dig into that next.
> 
> ******************************************************************************
> 
> So then, it's not as fancy as presented. It uses an NVMe driver, and
> for Clonezilla, if Clonezilla can handle an NVMe drive, then it can
> handle that thing.
> 
> Normally, when we analyze these things, we extract info about the implementation,
> from the RAID modes supported. You'll notice for a lot of this crap, it
> never goes past RAID0/1/10. RAID5/6 requires XOR for read/write and the chip
> runs hot while doing so. The four drive version of the card is still limited
> on RAID modes (does not magically do RAID5). In the four drive card case,
> I/O rates per sled are cut in half (2*1GB/sec lanes).
> 
> The card in question is PCIe Rev3, which is on the order of 3500MB/sec or so.
> In RAID1 mirror mode, you're "looking at one stick". If it really does
> use all eight lanes in the PCIe switch, then with two JBOD drives it should
> be able to pull 7000MB/sec (like if you used a softraid on top of it maybe).
> 
>   https://www.digchip.com/datasheets/parts/datasheet/2/277/88NR2241B.php
> 
>     "No custom drivers are necessary as the accelerator supports native NVMe in-box drivers
>      and is compatible with most popular operating systems today such as VMware ESXi,
>      Windows Server and popular Linux distributions."
> 
> Another citation in the CoPilot here:
> 
>   https://filedn.com/lG3WBCwKGHT7yNuTsFCwXy0/HighPoint-Download/Document/Guide/SSD6200/HighPoint%20SSD62xx%20Series%20CLI%20User%20Guid.pdf
> 
> This suggests that unless otherwise stated, it won't boot on my 4930K X79 because
> that has no NVMe INT13 support, and I can only use NVMe as data drives seen at OS time.
> It would work on the newer Zen3 boxes, which all have NVMe sleds onboard. It might
> not be as good as some other card we discussed, that seemed to have a boot EEPROM
> on it to make that product visible to the BIOS.
> 
> At least CoPilot confirms that some of the marketing weasel words are not correct.
> It looks approximately like an NVMe stick.
> 
> Just a guess,
>     Paul


Hi Paul,

You put a lot of work into that!  Thank you!

You did a good job of trudging through the "AI Slop"

-T