RAID‑10
10+ years
$0five‑year SSD cost
- SSDs you buy
- 0
- SSDs you replace
- 0
- Usable, with your data
- 0 TB
maxRAID
Software RAID,
redesigned for SSDs.
RAID was built for spinning disks. maxRAID was built for flash: RAID‑10 speed with the efficiency of RAID‑5 and RAID‑6, longer drive life and inline compression — all at once.
The old trade‑off
Fast or efficient.
Pick one. Until now.
RAID‑10
Fast, but wasteful.
Half of every drive you buy goes to mirroring. You pay for it, power it, cool it — and only half reaches your applications.
RAID‑5 / RAID‑6
Efficient, but slow.
Every small write turns into a read‑modify‑write. Performance collapses under random writes, and SSDs wear out years early.
maxRAID
Fast and efficient.
Parity capacity with better‑than‑mirror speed, less wear and compression on top. One change in how data is written.
Performance
Faster than RAID‑10.
With parity.
Parity RAID that is faster than mirroring in our tests.
- Predictable as it fills. Performance stays predictable as the array fills — keep some free space for the best results. We recommend keeping drives about 80% full.
- More parity, no penalty. 3‑ and 4‑drive parity with minimal performance impact.
The idea
Stop rewriting. Start streaming.
maxRAID places a flash translation layer above the RAID layer. Writes are gathered, compressed and aligned, then committed as full stripes. The read‑modify‑write penalty is avoided.
Traditional parity RAID
One small write. Four I/Os.
- Read old data
- Read old parity
- Write new data
- Write new parity
DDDDP
Every random write becomes a read‑modify‑write. Throughput collapses and the parity drive is hammered.
maxRAID
Many small writes. One stripe.
Writes gather in memory, compressed and aligned
DDDDP
Parity is computed in memory. The stripe is written once, in full, with no reads. The SSDs only ever see sequential writes.
Efficiency
Fewer drives
for the same capacity.
120 TB of usable storage on 15 TB SSDs. RAID‑10 mirrors everything, so it needs sixteen drives. maxRAID uses parity — and compresses what it can.3
RAID‑10
16
maxRAIDno compression
9
maxRAIDtypical data
7
Fewer drives. Even before compression.
With highly compressible data, as few as 5.
Less to buy.
Fewer drives, smaller servers, less rack space. Savings start on day one.
Less to run.
Fewer drives to power, cool, monitor and replace — every year.
More protection.
Choose 1 to N‑1 parity drives. Add protection without giving up speed.
Endurance
Drives that last
years longer.
Traditional RAID‑5 writes every update twice — data and parity — so a workload that lasts three years on RAID‑10 can wear out RAID‑5 drives in under two. maxRAID writes less than RAID‑10 does.
- Less wear on every write. For each 1 GB you write, maxRAID‑5 writes 1.15 GB to the SSDs and maxRAID‑6 writes 1.33 GB. RAID‑10 writes 2 GB.
- Within 0.5% of the optimal model. Real‑world write amplification, validated against Monte Carlo simulations since 2015.
- Cheaper media, safely. With far less wear, QLC SSDs may handle workloads that typically require TLC.
- RAID‑5 20 months
- RAID‑10 36 months
- maxRAID 5+ years
No drive swaps. No maintenance windows.
How fast does RAID‑5 wear SSDs?
RAID‑5 life ≈ RAID‑0 life ÷ 2 × N ÷ (N − 1)
36 months on RAID‑0 becomes about 20.5 months in an 8‑drive RAID‑5 array.
What will your SSDs cost over five years?
Three questions. We’ll compare RAID‑10, traditional RAID and maxRAID: how long the drives last, and what they cost.
Over five years, maxRAID costs less in SSDs.
10 × 15.36 TB SSDs: 8 data + 2 parity · drive life5 and cost over five years, drives only
Traditional RAID‑6
10+ years
$0five‑year SSD cost
- SSDs you buy
- 0
- SSDs you replace
- 0
- Usable, with your data
- 0 TB
maxRAID
10+ years
$0five‑year SSD cost
- SSDs you buy
- 0
- SSDs you replace
- 0
- Usable, with your data
- 0 TB
How it’s calculated
- Data written per day
- 61 TB
- Writes each can sustain for 5 years
- 2.10 vs 0.42 array writes/day
- maxRAID vs traditional
- maxRAID write amplification
- 1.59
- flash effectively 63% full
- Flash endurance used
- 3,000 P/E cycles
- traditional RAID uses the 1 DWPD rating
- SSD’s internal spare flash
- 28%
- Not directly accessible to maxRAID — the drive’s controller uses it to spread wear
- maxRAID free‑space reserve
- 10%
- Set above
Under the hood
Everything flash needs.
Nothing it doesn’t.
Write linearization.
Random writes become large, aligned sequential writes across every SSD.
Near‑theoretical garbage collection.
A host‑side FTL that minimizes write amplification and keeps performance flat.
Inline compression.
Blocks are compressed before they’re written. Often, write amplification drops below 1:1.
Automatic trim.
Blocks of all zeros or ones are discarded automatically — no client or hypervisor support needed.
Flexible parity.
From single to N‑1 parity, including 3‑ and 4‑drive protection, with minimal overhead.
Any SSD.
QLC, TLC, MLC or SLC. SATA, SAS or NVMe. Even drives with large indirection units.
Specifications
Drop it in. It just works.
maxRAID is a Linux kernel module that presents a standard block device — usable anywhere a block device is.
- Form
- Linux kernel module, standard block device
- RAID levels
- Single SSD, RAID‑0, RAID‑1, RAID‑5, RAID‑6, plus 3‑ and 4‑drive parity
- Linux
- RHEL 8 and 9 (plus select Rocky and AlmaLinux releases); Ubuntu LTS 22.04 and 24.04; custom kernels for OEMs and cloud providers
- Interfaces
- SATA, SAS, NVMe; fabric targets such as iSCSI and NVMe‑oF
- Media
- QLC, TLC, MLC, SLC; data‑center and consumer SSDs; large indirection‑unit drives
- Block size
- Presents 512 B or 4K; can issue 4K‑only I/O to drives that require it
- Works with
- File systems, LVM, partitions, swap (swap‑safe), TRIM/discard
- Snapshots
- Via LVM or snapshot‑capable file systems on top of maxRAID
- Export
- iSCSI or NVMe‑oF — or manage everything with maxRAIDOne
- Best for
- Moderate to high write workloads (≥ 0.3 drive writes per day) where performance matters
See it on your own hardware.
Try maxRAID free for 120 days, on arrays up to 100 TB.
- Measured: 4K random writes on six 3.75 TB NVMe SSDs. RAID‑5 and maxRAID‑5 use five data drives and one parity drive.
- Measured: 70% reads and 30% writes with 4K blocks, on six NVMe SSDs. maxRAID‑6 delivered 10.4 GB/s, or 3.1 million IOPS.
- Modeled: 120 TB usable on 15 TB SSDs, maxRAID with one parity drive. Typical data compresses by 30%; highly compressible data by 60%. Results depend on how well your data compresses.
- Modeled: 120 TB usable on 15 TB SSDs under the same heavy‑write workload. maxRAID needs no drive replacements within five years.
- Estimates from WildFire’s SSD wear model. Traditional RAID‑5/6 uses the drives’ rated writes per day over five years, with every write also written to parity. maxRAID uses the flash’s rated P/E cycles, including the SSD’s internal spare flash (over‑provisioning, 28% by default — used by the drive’s controller to spread wear, not directly accessible to maxRAID), the free‑space reserve you choose (10% by default), how full you keep the array (80% recommended), compression and measured write amplification. Workloads are array writes per day: light 0.1, moderate 0.3, busy 0.5, very busy 1. Prices are planning estimates; edit them to match your quote. RAID‑10 needs twice the drives for the same usable capacity and writes every block twice, using the same drive rating. Five‑year SSD cost counts drives bought plus drives replaced, at your price; it excludes servers and software. Actual results depend on workload and drive model.