Whats the difference between software RAID vs hardware RAID now in 2025?
I have always used software RAID when I opted to use RAID over running scheduled backups. Back in the dark ages of 2010s, software RAID was slow and horrible. A single issue with disk or motherboard could result in major data loss where the RAID disk is no longer recognized. JBOD used to be the best alternative to RAID 1 back then. As such, the recommended way to RAID would be with a hardware controller.
Now in 2025, is this still the case? Anyone who's using a hardware RAID controller from the last 3 years, could you share your experience vs using software RAID?
And for those who are using software RAID (I guess would be many of you) how is the data recovery process once one of the drives or the motherboard goes down? What is the process of getting your RAID recognized by the new hardware or the process of data recovery?
Finally, those who are still using scheduled/manual backups, are you planning to switch to RAID anytime soon?
FYI, backup = disk backup (or clone). As pointed out by the first post, RAID is not backup, so I would still be backing up my data, just not the entire disk image.
Never make the same mistake twice. There are so many new ones to make.
It’s OK if you disagree with me. I can’t force you to be right.
Comments
I know this doesn't answer your question, but I think most people would tell you that RAID is not a replacement for doing backups.
Ah, sorry for phrasing it incorrectly. What I meant by backup is "backing up the entire os image" instead of just my project/data.
Never make the same mistake twice. There are so many new ones to make.
It’s OK if you disagree with me. I can’t force you to be right.
Oh, phew. Glad it was just a misunderstanding
Oh yes, I am someone who keeps 1 backup on a different disk, 1 backup on a backup server in different country and finally 1 backup made manually on a offline portable disk. Losing your data is the worst...
I am also making a disk image backup on the same proxmox server so was planning to replace that with a RAID.
Never make the same mistake twice. There are so many new ones to make.
It’s OK if you disagree with me. I can’t force you to be right.
Honestly I've used linux mdadm raid with NVMe forever now, once the drive is swapped it literally takes seconds to run the commands to re-add a new drive and resync. I've replaced many failed drive in mdadm raid, never a problem. It's even better when you have remote hands who are super quick, downtime can only be a few minutes max.
As long as you have the original OS install/raid setup properly on both drives, it's great.
ExtraVM - KVM NVMe VPS in USA, EU, APAC -|- RackColo - Find Colo
Discord: mikea
Honestly I was using it back in 2012ish on a bunch of 36gig scsi's and it worked fine. Rebuilds weren't instantaneous but weren't problematic either. Just my opinion though.
That's great to know that rebuilding doesn't take forever anymore. May I check on the size of the RAID disks you have?
You guys clearly have better experience with RAID that I did back in the day. Then again, I was running it on Windows Server, not linux...
Never make the same mistake twice. There are so many new ones to make.
It’s OK if you disagree with me. I can’t force you to be right.
is this RAID 1 or 10?
I bench YABS 24/7/365 unless it's a leap year.
I only use RAID-1 with 2x M.2 NVMe, but the procedure is the same for RAID-10.
ExtraVM - KVM NVMe VPS in USA, EU, APAC -|- RackColo - Find Colo
Discord: mikea
Raid 1 = 2 disks,
Raid 10 = 4 disks, 2 on raid 0, then the pair on raid 1.
Sounds the same to me...
Never make the same mistake twice. There are so many new ones to make.
It’s OK if you disagree with me. I can’t force you to be right.
You are one sick puppy.....
Free Hosting at YetiNode | MicroNode| Cryptid Security | URL Shortener | LaunchVPS | ExtraVM | Host-C | In the Node, or Out of the Loop?
I’ve found Debian to be super finicky with setup/install to achieve it. Annoying at times to ensure EFI is synced etc.
Any tips or pointers would be hugely helpful!!
Rock Solid Web Hosting, VPS & VDS with a Refreshing Approach - Xeon Scalable Gold, DDoS protection and Enterprise Hardware! HostBilby Inc.
No idea, I only use Almalinux/RHEL OS.
ExtraVM - KVM NVMe VPS in USA, EU, APAC -|- RackColo - Find Colo
Discord: mikea
Hey, that was back in the day of Windows Server 2012... And set it up for my company which had some windows server software which was written in .net for windows...
Never make the same mistake twice. There are so many new ones to make.
It’s OK if you disagree with me. I can’t force you to be right.
Did you follow this guide?
https://bmcservers.com/how-to-create-raid-ubuntu24
It uses mdadm that @MikeA mentioned but it's for ubuntu, so should be applicable for debian as well.
Never make the same mistake twice. There are so many new ones to make.
It’s OK if you disagree with me. I can’t force you to be right.
Recommend just creating mdadm raid in the OS installer in the first place :0
ExtraVM - KVM NVMe VPS in USA, EU, APAC -|- RackColo - Find Colo
Discord: mikea
sounds interesting. i might be ready to play with dedicated server soon™
I bench YABS 24/7/365 unless it's a leap year.
+1 to this for software raid, mdadm is what I on Debian, can be setup in 60 seconds
It's also the default that OVH and a few other providers use for bare metal raid templates on deb iirc
lex.st - Free Shared Hosting in 4 Locations (KC USA now live - sponsored by Incognet)
Why do i have 10 old HDDs for playing around with but all are of different capacities?!? Oh well, at least according to the internet, it'll still let me create a RAID, just that i'll take the size of the smallest device. Oh well, will update once I find out.
So basically what you are saying is most no longer use dedicated RAID controllers and just go with software RAID since the performance is almost the same? We really have come a long way since those dark days...
Never make the same mistake twice. There are so many new ones to make.
It’s OK if you disagree with me. I can’t force you to be right.
Just look at the lowendX deals for VPS. Most VPS with Ryzen are running raid-1 NVME drives in software raid, you can get great performance. Plus software raid removed a point of failure that hardware raid has... extra hardware.. just look at the history of complaints on WHT 10 years ago about hardware raid cards failing and the whole system getting fked. But also, hardware in general is insane compared to 10+ years ago too so software raid probably has no real bottleneck now.
ExtraVM - KVM NVMe VPS in USA, EU, APAC -|- RackColo - Find Colo
Discord: mikea
That's what I was saying in my first post. Raid controller spoin = whole RAID gone...
Yes, the hardware did improve but I was stuck on the following mindset:
However if we consider the current multi disk servers we have, writing to 1 disk does not affect the other disks. So the only issue is calculating the parity, but that's nothing with our current CPUs. So this makes more sense (do correct me if I'm wrong):
So with RAID with 4 disks (as per the link on the site):
Never make the same mistake twice. There are so many new ones to make.
It’s OK if you disagree with me. I can’t force you to be right.
Is anyone even doing hardware RAID anymore? I get the sense that everyone moved to software because it is less of a mess on hardware failure (specifically the controller)
I do because of overhead and it gives me dedicated control. Also I am old skool and hate change.
Free Hosting at YetiNode | MicroNode| Cryptid Security | URL Shortener | LaunchVPS | ExtraVM | Host-C | In the Node, or Out of the Loop?
Just some random nodes.
Now, here is my opinion on this. There are 2 perspectives on the question, HW or SW, and that is use case.
Home use / Personal Use, SW is fine as hell, up to 4-8 drives it is perfect, you can even run a small number of VM's from a raid 6 setup like that, store stuff, perfect. Raid 6/Z2 will give you the benefit of any 2 drives can fail, that is cool, while over a 1 GBPS ETH line you will not need anymore. For High IO go with Raid 10.
Professional/Hosting and high number of drives ( over 12 )
Now here things start to change a lot.
SW raid Raid 6/z2 will consume CPU ( light for MDADM and heavy for ZFS ) and a ton of RAM for ZFS. Also, you will use the CPU to do the math on Parity Blocks, and the same CPU has to do other functions, like OS emulation and provide VPS speed. So waiting CPU Cycles on that is..... not wise.
Secondly, if the location is colocated, to change drive SDF will be a pain in the ass, as you will most probably not have the RED diagnostic LED on as the BAY status is not controlled anymore by the HBA ( 90% of the cases ) and debugging to unplug the correct drive can take a lot of hours. So for a provider, it just adds complexity to a problem that is rather a qestion of when then if it will happen. Drives fail all the time, for example we currently change ~3% of drives each year due to failure.
HW raid with a modern controller will bring some benefits, first, Latency is minimal, no CPU cycles wasted on storage and no RAM consumed by the host.
Secondly it will light up the LED on the Drive failed making replacement a breeze ( unless the remote hands on in the DC is color blind, and unplugs the wrong drive, not that it ever happened....
) but even then it will not be a problem and things can be remediated fast.
On large number of drives we use RAID60 ( 2xR6 of12 drives ) as it performs much better IO wise. We would use RIAD10 and achieve near NVME speeds on 24 drive setups while burning 50% of capacity, but I doubt anyone will pay 6USD / TB / MO.
So for a Provider stand of view, I only see benefits going HW raid, there is a catch however, if the HW raid controller is not something from this era, it can make things very very bad. So spending a some serous USD on it makes total sense.
What not to do:
RAID5 / RAIDZ1 - since we passed the 2/4 TB / Drive mark, you have a 50% of another failure within 24 hours when you replace a failed drive. Since raid5/Z1 has only 1 parity disk, you have 50% of loosing your porn-hub collection.
Raid5/Z1 or other 1 parity drive setups are not used in Production for over a decade. If yu have 4 bays only, just do raid10 as that is the best.
Having 2 field drives on the same span in a raid 10 is extremely unlikely, plus the rebuild on raid 10 ( stripped/added mirrors ) does not imply any math, just pure copy from the healthy member, so even on a 4x 20TB raid setup, rebuild will be extremely fast and IO delay almost none, regardless if it is HW or SW raid.
Host-C - VPS & Storage VPS Services – Reliable, Scalable and Fast - AS211462
"If there is no struggle there is no progress"
I agree 100% with you on the failing HW raid Controllers 10 years ago, especially since most used HP servers, and the P400/P410 cards had a tiny heat-sink and usually 85Celsius working temperatures. I have seen so many felled off heat-sinks on HP that I cannot say.
LSI based cards were a Joke before SAS3 Standards, so yes, at that point SW could actually beat some low-mid ranged HW raid cards. ( I used ZFS back then and beat the shit out of High End Raid Cards, even latency wise in RID10 as I did not need to do parity calculations )
Today things changed ( well I would put ~5 years ago ), a lot, modern CPU's on high end raid cards run ~50-65 Celsius and they carry dedicated ASIC for Parity Calculation, that are not easy to do in large setups, so they are extremly low in latencye.
Most modern cards have 8GB and above Cache, that is extremely fast and are ECC.
Software on the Cards got bigger and "smarter" as they now carry much more powerful CPU's.
These are light-years ahead of what was common 10 years ago.
Parity RAID needs CPU cycles for XOR/parity math. On high-IO or high-density storage servers, that steals performance from your applications unless you’re massively over-provisioned.
HW RAID offloads that entirely with near-zero latency and sustained throughput.
As I said before, it really depends on the use case wherever to go SW or HW.
It is like using the right tool for the right job.
Host-C - VPS & Storage VPS Services – Reliable, Scalable and Fast - AS211462
"If there is no struggle there is no progress"
I’m using mdadm currently to sync OS drives and use ZFS mirrored vdevs for prod data.
But Debian doesn’t make it easy to setup drives in software raid as part of install, along with handling efi and boot etc… I also haven’t gotten around to figuring out preseed to make things scalable yet.
I think TenantOS handles it well - so I’ll have to poke around at the preseed drive portion and see what I’m missing!
Rock Solid Web Hosting, VPS & VDS with a Refreshing Approach - Xeon Scalable Gold, DDoS protection and Enterprise Hardware! HostBilby Inc.
You can setup mdadm raid in the Debian graphical installer for everything but it's not user friendly, I just pulled up the installer to check. I never use Debian since my whole business runs on Almalinux but I remember the installer being a pain to work with.
ExtraVM - KVM NVMe VPS in USA, EU, APAC -|- RackColo - Find Colo
Discord: mikea
That looks like half a petabyte of drives? o_O
Nice info! Thanks for taking the time
Anybody know how this works at hyperscaler level? I heard google does triple region replication for gmail data so guessing it flips back to software at that level
Debian/ubuntu gui installer needs a lot more work... I mostly use ubuntu server gui installer but I dont remember seeing the mdadm raid option... I guess we can always live boot the OS and setup mdadm and then reboot to start the installation but not very fast or user friendly. Do note that I haven't tried this yet. Will give debian installer a try as well.
Now that I think about it, I cant recall why I started using ubuntu over debian...
Never make the same mistake twice. There are so many new ones to make.
It’s OK if you disagree with me. I can’t force you to be right.
Had exactly this issue some years ago, the guy changed the wrong disk and put one with different RPM 🙃
Free server monitoring → fivenines.io