The first meeting
I sent the maxRAID salesperson away in under twenty minutes. For a year, I let that claim sit in the back of my mind while I kept buying hardware. Then I finally ran the test. Six months after that — having fully realized what those numbers meant — I left the data center I’d spent my career building to go to work for the company that makes it.
He’d told me I could have the responsiveness of RAID‑10, use my drives more efficiently, write less to them so they’d last longer, and get compression on top — all at once, from a single change. I know what it means when everything in a pitch improves at the same time. It means a bill you haven’t found yet. Performance, capacity, endurance, cost — you trade among them, and you don’t get to stop trading. So I showed him the door.
The compromise
My customers never saw a single one of my storage decisions. They saw the result: they knew when an application answered the instant they asked, and when it made them wait. So I built for response, and for years, that meant RAID‑10.
I also knew exactly what it cost me. Roughly half of every array I bought went to mirroring. I paid for every drive, powered it, cooled it, monitored it, and replaced it when it wore out — and only half of what I’d paid for ever reached a customer. The answer was always more hardware.
The RAID‑10 tax
The year spent deciding instead of measuring
I came back to it for a reason I’m not proud of. It just wouldn’t leave. It surfaced at the worst possible moments — usually while I was approving the next hardware purchase.
Eventually, I had to be honest about what I’d actually done. Dismissing a claim without testing it isn’t skepticism. It’s the opposite — it’s taking someone’s word for it, only in the other direction.
A real engineer doesn’t believe a vendor. A real engineer doesn’t disbelieve one either. A real engineer measures.
It stopped being magic
It sounds like magic because it’s described as four outcomes — faster, leaner, longer‑lasting, compressed. But it isn’t four things. It’s one, and the four wins are only its shadow.
maxRAID places the translation layer above the RAID rather than below it. Scattered, random writes are gathered into long, sequential ones and compressed on the way down to the drives. That single move is the cause. Everything else follows from it without anything being added.
The line I wouldn’t cross
At Halton, power, network and cooling are all redundant. Storage has to obey the same rule. RAID‑5 survives one dead drive, but the moment it does, its protection is spent. RAID‑6 stays protected after the first failure — but for my entire career, RAID‑6 has been charged for that protection in the one currency I couldn’t spare: response.
| Protection, compared | RAID‑5 | RAID‑6 | RAID‑6 + maxRAID |
|---|---|---|---|
| Survives a drive failure | Yes | Yes | Yes |
| Still protected after that failure | Exposed | Protected | Protected |
| Response under load | Slow | Slower | RAID‑10‑class |
Then the numbers came back
I built the test myself, on my own hardware, against the workloads my customers actually generate. No vendor in the room, no demo rig. I read the results once. Then I read them again, slower, looking for the catch. It wasn’t there.
Measured at Halton Data Center
6 × 3.75 TB NVMe · maxRAID dual parity (RAID‑6) · 30% compressible workload
19,941MB/s
Throughput — RAID‑10‑class
2.46M IOPS
70/30 at 4K, sustained
2.86M IOPS
100% writes at 4K, sustained
21.43TB
Effective capacity, vs 11.25 TB on RAID‑10
RAID‑6 protection. RAID‑10 speed.
More to sell. Same six drives.
Same drives. Same rack, same power, same cooling. RAID‑6 protection. RAID‑10 response. Less wear. More capacity. The compromise I had organized fifteen years of decisions around was simply gone, and nothing anywhere was paying for it. The drives weren’t working harder. They were finally working right.
It decided the rest of my career
I didn’t just want to fix the storage efficiency inside my own data center. I wanted to be part of the team fixing it for every operator who was still paying the same hardware tax I had tolerated for fifteen years.
If you know too much to believe this — good. So did I, and that instinct is worth keeping. But don’t take an afternoon on faith. Take it on falsifiability. If the numbers don’t show up on your hardware, you’ve confirmed your instinct and lost nothing. If they do, you’ll be standing exactly where I stood.