Latest / Big Iron Radio / RAID Showdown: Hardware RAID vs ZFS vs Btrfs vs mdadm

Transcript

RAID Showdown: Hardware RAID vs ZFS vs Btrfs vs mdadm · Big Iron Radio

• 0:03
  Sooner or later, every homelab hits the same milestone.
• 0:07
  You outgrow a single drive, and you decide to build an array.
• 0:11
  And the moment you start reading, you get pulled into what looks like a performance
• 0:15
  argument.
• 0:17
  Hardware RAID vs. ZFS vs. BTRFS vs. MDATUM which one's fastest?
• 0:22
  Which one's easiest?
• 0:23
  Today, I want to convince you that speed is the wrong axis to start on.
• 0:29
  The real question is whether your storage can even tell you when it's lying to you.
• 0:34
  Quick note before we dig in.
• 0:36
  This is an AI-generated voice, and everything here is adapted from a written guide over at
• 0:42
  Big Iron.
• 0:43
  The full write-up with the RAID's layout breakdown and a rebuild risk calculator is linked in
• 0:48
  the show notes.
• 0:50
  So four contenders.
• 0:53
  Hardware RAID, a dedicated card that hides the disks from your operating system.
• 0:57
  MDATUM, the plain Linux software RAID that's been in the kernel forever.
• 1:02
  BTRFS, a modern copy-on-write file system with checksums baked in.
• 1:08
  And ZFS, the integrity gold standard from the storage world.
• 1:12
  These are not four flavors of the same thing.
• 1:15
  They sit in genuinely different places on the axes of performance, integrity, flexibility,
• 1:21
  and operational maturity, and picking wrong can get you data loss you can't even detect.
• 1:26
  Let's talk about that phrase data loss you can't detect, because it's the whole ballgame.
• 1:32
  It's called silent corruption, or bitch-rot.
• 1:35
  A bit on the platter, or in the flash, quietly flips.
• 1:38
  Cosmic ray, aging cell, a firmware hiccup doesn't matter.
• 1:42
  The drive doesn't throw an error.
• 1:45
  It hands that slightly wrong data back to you as if nothing happened, and it works its way
• 1:49
  into your files, and, eventually, into your backups.
• 1:54
  Here's the uncomfortable part.
• 1:56
  Classic RAID does nothing about this.
• 1:59
  RAID gives you redundancy a second copy, or parody to rebuild from.
• 2:03
  But when it reads a block, it has no way of knowing that block is wrong.
• 2:08
  There's no fingerprint to check it against.
• 2:11
  Hardware RAID especially just hands the operating system bytes.
• 2:15
  If those bytes are silently wrong, no layer above it ever catches on.
• 2:20
  Silent bit-rot on Hardware RAID is the classic home lab disaster the array is healthy, green
• 2:25
  lights all around, and your photos are quietly rotting.
• 2:30
  This is exactly what a checksumming file system fixes, and it's the single most important
• 2:35
  idea in this whole comparison.
• 2:37
  ZFS and BTRFS checksum every block.
• 2:41
  When they read a block whose checksum doesn't match, that's not a shrug it triggers a reread
• 2:46
  from a redundant copy, the mirror partner, or a parody reconstruction, and the bad block
• 2:51
  gets repaired.
• 2:53
  Silent corruption becomes loud.
• 2:55
  Once that clicks, the phrase RAID without checksums is only half a solution,
• 2:59
  stops being a slogan, and starts being obvious.
• 3:03
  OK the contenders, one at a time.
• 3:07
  Hardware RAID first.
• 3:08
  This is a dedicated PCIe card with its own processor, its own battery-backed write cache,
• 3:14
  and proprietary firmware.
• 3:16
  It juggles your drives behind the scenes and presents the OS with one big block device.
• 3:22
  On paper it looks great.
• 3:24
  Along sequential write throughput, parity offloaded to the card, and its operating system agnostic
• 3:30
  the array doesn't care if you're on Windows or Linux.
• 3:34
  But the weaknesses are serious.
• 3:37
  No file system level integrity, as we just covered the card can't catch silent corruption,
• 3:43
  and its patrol read only checks that parity is consistent, not that the data is actually
• 3:48
  correct.
• 3:49
  There's vendor lock-in.
• 3:51
  If the card dies, you often need the same card family to even read your array.
• 3:56
  And there's the battery.
• 3:58
  Those backup batteries die after two to four years, and without a working one, the write-back
• 4:03
  cache is no longer safe so the card quietly downgrades to write through, and the performance
• 4:08
  you paid for evaporates.
• 4:11
  So the honest verdict on Hardware Raid, its era is mostly over for new builds.
• 4:17
  If you've inherited an array on a good card, there's no need to panic and rip it out.
• 4:22
  But if you're building fresh in 2026, you almost certainly shouldn't be buying into it.
• 4:28
  The move instead is an HBA, a host bus adapter, flashed to what's called IT mode, which just
• 4:34
  exposes the raw disks, and lets a smart file system do the thinking.
• 4:39
  Hold that thought.
• 4:41
  Second contender, mdatum.
• 4:44
  This is vanilla Linux software raid, straight out of the kernel.
• 4:48
  It's mature, it's rock solid, and it composes beautifully with the rest of the Linux stack
• 4:53
  you can layer LVM on top, put encryption underneath, mix and match.
• 4:59
  If you want boring, well understood, no surprises software raid, mdatum is it.
• 5:05
  Its limitation is the same integrity gap.
• 5:08
  Mdatum doesn't checksum your data, so it can't catch silent corruption on its own.
• 5:13
  And its parity modes carry the classic RAID 5 slash 6 write hole, and interrupted write
• 5:18
  can leave a stripe inconsistent.
• 5:21
  So mdatum is the right pick in one specific situation.
• 5:25
  When your data integrity is coming from somewhere else full disk encryption with authentication,
• 5:31
  your application level checksums, and you just want straight forward kernel RAID underneath.
• 5:36
  And to kill an old myth, modern CPUs handle RAID 5 slash 6 parity without breaking a sweat.
• 5:44
  Software RAID is slow is a reputation from decades ago.
• 5:48
  Third, BTRFS.
• 5:52
  Now we're into checksumming territory.
• 5:54
  BTRFS is a copy-on-write filesystem with checksums, instant snapshots, and reflinks,
• 6:00
  and crucially it's in the mainline Linux kernel no out-of-tree modules, no install friction.
• 6:06
  For a Linux root filesystem, where you want snapshots before every upgrade, BTRFS is genuinely lovely.
• 6:14
  But BTRFS comes with one caveat you must respect.
• 6:18
  Its RAID 1 and RAID 10 modes are great and fully checksummed, but its RAID 5 and RAID 6
• 6:24
  are not safe for production.
• 6:26
  The parity write hole was never fully resolved there.
• 6:30
  So use BTRFS for mirroring, use it for your root filesystem, use it where checksumming matters
• 6:36
  and you specifically don't want ZFS's licensing and install friction, but do not trust BTRFS
• 6:42
  RAID 5 slash 6 with data you care about.
• 6:46
  That's not caution, that's the project's own guidance.
• 6:50
  And fourth, the one the storage nerds evangelize, ZFS.
• 6:55
  It's the integrity gold standard, and it earns the title.
• 6:59
  Copy-on-write end-to-end checksums, its own RAID levels called RAIDs, native snapshots,
• 7:05
  transparent compression, built-in encryption.
• 7:07
  It was designed from the ground up around the assumption that hardware lies, and its whole
• 7:12
  job is to catch the lie.
• 7:15
  The feature that shows off the philosophy is the scrub.
• 7:18
  You run a scrub on a schedule, and ZFS walks every block, verifies every checksum, and where
• 7:24
  it finds corruption it doesn't just flag it it identifies which copy is wrong and repairs
• 7:29
  it from the good one.
• 7:31
  Compare that to a hardware card's consistency check, which can only tell you parity and data
• 7:36
  disagree, not which one to believe.
• 7:39
  That difference detect vs. detect and correctly repair is the entire reason people put up
• 7:44
  with ZFS's quirks.
• 7:47
  And there are quirks.
• 7:49
  ZFS is licensed under the CDDL.
• 7:52
  So on Linux, it lives outside the mainline kernel in extra install step and a version to keep
• 7:57
  an eye on.
• 7:58
  It's RAM-hungry by design.
• 8:01
  And pool expansion is more constrained than people expect you can't always just slot in
• 8:06
  one more disk, the way you imagine so you plan your VDEV layout up front.
• 8:11
  That planning is where RAIDs 1 vs. RAIDs 2 vs. Mirrors comes in.
• 8:16
  Mirrors for maximum IOPS and easy growth, RAIDs 2 for capacity with 2-disc fault tolerance.
• 8:22
  There's a calculator in the show notes to weigh the rebuild risk.
• 8:27
  One pairing you can't skip with ZFS, and it ties back to another big iron brief is memory.
• 8:33
  ZFS protects the path to the disk, not the path through your RAM.
• 8:38
  Corrupt memory still corrupts data before ZFS ever checksums it.
• 8:43
  So the serious build is ZFS plus ECC RAM, plus regular scrubs, plus smart monitoring on
• 8:49
  the drives.
• 8:51
  Integrity is a stack, not a single feature.
• 8:54
  And that HBA point from earlier closes the loop.
• 8:58
  If you go ZFS or BTRFS, don't sit them on top of a hardware RAID card.
• 9:04
  Give them an HBA and IT mode so the filesystem sees the raw disks, and its checksums actually
• 9:09
  mean something.
• 9:10
  A smart filesystem on top of a card that's already hiding the disks is the worst of both worlds.
• 9:18
  So let's turn all of that into a decision you can actually make.
• 9:21
  If data integrity is your priority, and you're on Linux, BSD, Illumos, or Mac OS, and you
• 9:28
  can pair it with ECC ZFS, it's worth the setup friction.
• 9:33
  That's the default for anyone who thinks of their storage as the thing they'd cry about
• 9:37
  losing.
• 9:39
  If you want checksums on Linux, but you don't want ZFS's licensing and install friction,
• 9:44
  BTRFS on RAID 1 or RAID 10 is an excellent choice just stay off its RAID 5 slash 6.
• 9:51
  If your integrity is handled at another layer, and you want plain, mature kernel RAID, Mdatum
• 9:57
  is a perfectly respectable answer.
• 10:00
  And hardware RAID, reserve it for legacy systems, vendor-locked appliances, and Windows-only environments
• 10:07
  not new builds.
• 10:09
  The one thing to carry out of this, rank integrity above raw performance.
• 10:14
  A fast array that silently corrupts data isn't fast, it's just quick at losing your files
• 10:20
  in a way you won't notice until restore day.
• 10:23
  Checksums are what turn a pile of redundant disks into storage you can actually trust.
• 10:29
  The full guide, the RAID's layout deep dive, and the rebuild risk calculator are all in
• 10:33
  the show notes.
• 10:35
  Thanks for listening, I'll catch you on the next one.