Debating between rebuild of [current] HPE Array Hardware RAID or going with OMV Software RAID

  • Hello


    Either way, I need to rebuild my Array as I currently have 9 2.73 TB drives in a RAID (5(don't quote)) config but am wanting to incorporate the remaining 3 so total of 12 2.73 TB drives.

    I know things have changed and I hear so many pros and cons to RAID 5 and I am totally open for discussion. My main questyion is, do I continue to use the HPE Proliant DL380 GEN10 RAID Controller card, which thus far have been happy with, or being that this system is and will only be for OMV, do I go Software RAID?

    Ideally I want the most capacity of whatever RAID I choose, but I am absolutely fine losing capacity for security ( I mean, 8+/- TB is already not part of the RAID so) I just do not know. I believe HPE offers RAID0, RAID ADG, RAID5 and RAID1+0.

    I just do not know! I see so many pros and cons, that I wanna ask people who ave tried.

  • I'm gonna say a few thing about RAID levels and hardware vs. software in general, and explain my choice to use a RAID5 while I still recommend that most people don't need it.


    1.) RAID is not a backup. RAIDs can still fail and cause data loss. Always backup your important data too.


    2.) RAID in simplest terms a way to pool drives together, either into a bigger volume, and/or with some protection against drive failure, and/or to increase the performance compared to a single HDD that the RAID is built from. (I reference HDD, but RAID can also be built from SSD or NVME, however the performance benefit is not as noticeable to the average user due to the already enhanced performance of solid state storage. Solid state RAID performance is noticeable more so when the simultaneous user access numbers increase more or the data bandwidth needs increase beyond what you can reasonably achieve from a large array of HDD's)


    3.) Common RAID levels are 0, 1, 3, 5, 6, 10 (or 0+1), 50, and 60.

    RAID 0 is a data stripe across multiple drives with no protection. It offers the best performance, but if one drive dies, all data is lost.

    RAID 1 is a mirror and the capacity of one drive is lost for the mirror, (ie. drive 1 is mirrored onto drive 2, or there can be more than one mirror) There is no performance benefit over a single drive. If a drive dies, the data still exists on drive 2, but until that drive is replaced and the RAID is rebuilt, there is no protection.

    RAID 3 an 5 are essentially a RAID 0 with an extra drive used for a CRC check. 1 drive capacity is lost for the CRC, but the performance is close to RAID 0 (writes do tend to suffer due to the CRC calculations) The 3 and 5 levels just differ in if there is a dedicated CRC drive (3) or a distributed CRC (5)

    RAID 6 is like RAID 5 but uses 2 CRC's, looses the capacity of 2 drives to CRC and can suffer 2 drives failing before data loss

    RAID 10 or 0+1 is either a mirror across RAID 0 stripes, or a RAID 0 stripe across mirrors, the difference is the order in which they are applied

    RAID 50 and 60 are a RAID 0 stripe across multiple RAID 5's or RAID 6's. Each block of RAID 5 or 6 has the same features as a single RAID 5 or 6 but the RAID 0 stripe aims to improve the performance.


    4.) The more complicated the RAID the more it will benefit from a hardware RAID controller to take the CRC calculations off of the system CPU, and generally, it is easier to replace a failed drive on a hardware based RAID because the controller automatically looks after initiating the rebuild when it detects a new drive being inserted into a slot it has already marked as failed. Software mdadm RAID usually requires you to mark the drive as failed, then remore and replace it, then add the new drive back to the array to trigger the rebuild.


    5.) Generally, It is possible to add additional drives to a software RAID, but not a Hardware RAID. However I would never attempt it without a current full data backup first, in case things go sideways.


    6.) RAID 1 really has no performance benefit and almost the same drive failure safety can be achieved by regular backups to another drive (rsync, borg backup, etc.) Unless you need the added performance that a RAID 0, 3, 5, 10, 50, or 60 offers, you are probably safer using SNAPRAID or MergeFS to pool the drives. MergeFS adds the drives to a pool with one volume, but makes them operate as individual drives under the hood and SNAPRAID does the same thing, but adds some CRC checking to allow for up to 2 drives to fail, if I am not mistaken, before data loss.


    If it is purely the largest volume you are after MergeFS or SNAPRAID would be my recommendation, as they pose less risk of total data loss than a RAID does.


    I use a RAID 5 with an XFS file system because I needed the benefit from the added performance for simultaneously streaming multiple high bandwidth video files from from it for editing of broadcast and film level projects I do, and the projects can be several TB in size so the size of the volume is also important. I chose XFS as the file system because it is a very mature file system, it tends to work better with larger static files like large video assets than ext4 does, it offers the ability to defrag a live filesystem (ext4 supposedly now has this with the e4defrag tool also, but I have never tried it), which is important for streaming those large, high bandwidth video files, and like most, it has a fairly decent set of other tools for dealing with the file system.


    If I was not requiring those capabilities, I would likely not be using a RAID 5, but I would still probably use XFS

    Asrock B450M, AMD 5600G, 64GB RAM, 6 x 4TB RAID 5 array, 2 x 10TB RAID 1 array, 100GB SSD for OS, 1TB SSD for docker and VMs, 1TB external SSD for fsarchiver OS and docker data daily backups

    Edited 4 times, last by BernH ().

    • Official Post

    I personally like to use the MDADM soft RAID for RAID5/6 instead of HPE's own solution (and I work for HPE!)


    My reason behind it is that MDADM is not hardware dependent, while a failure on your HPE SmartArray controller is the complete opposite.


    As far as performance is concerned, I don't see any advantages other than the writing caching features of the HPE controller (which is dependent on a battery that needs regular servicing) but it also depends on how much data you will be writing on a regular basis and how many clients will be using it.


    As long as you are aware of the limitations of hardware RAID and are able to get spare parts at any time, I don't see any reason to move away from it.

    Absolutely no support through PMs!!
    I work for the HPE R&D division. Have tech questions about ProLiant servers? Ask me!
    Learn more about HPE's iLO included in all ProLiant servers.

Participate now!

Don’t have an account yet? Register yourself now and be a part of our community!