What the word SSD hides about your host's storage
Enterprise SSD on a pricing page can mean a consumer drive behind one RAID card or a dedicated dual-controller all-flash SAN, and only one of those outlives the server. Five questions, one email, and any host has to tell you which one they run.
In this article 5 sections
- Much of the storage sold as enterprise is consumer or prosumer SSDs behind a single RAID card, inside the same server that runs your workload. When that server dies, the storage dies with it.
- Ours is a SAN: dedicated dual-controller all-flash arrays, reached over two independent fabrics, serving at single-digit-millisecond or better.
- We standardized on one storage platform, HPE 3PAR all-flash over Fibre Channel, and mastered it instead of running a vendor menu.
- Five questions at the end of this post expose the difference in one email. Forward them to your host, and to us.
A hosting order page gives you exact numbers for cores and RAM and one word for storage, usually SSD. That one word is covering the widest quality gap in the industry. Two hosts can print the same word and sit an entire architecture apart: one keeps your data on a consumer drive behind a RAID card inside your server, the other keeps it on a dedicated all-flash array that goes on serving after that server dies. Here is the difference, plainly, and at the end, the five questions that will get the truth out of any host, including us.
One word, SSD, is covering three different architectures
Storage in hosting comes in three tiers, and the product copy for all three reads nearly the same.
- 01Local disk. Drives inside the server, on the server's own controller. Fast to provision and cheap to sell, and the storage shares every fate the box has: power supply, backplane, motherboard, controller. The whole tier is one failure domain.
- 02SSDs behind one RAID card. Where most hosting sold as enterprise actually sits. The drives are usually consumer or prosumer parts chosen by the gigabyte, and RAID protects you from exactly one thing, a dead drive, which is the cheapest failure storage has. It does nothing about a dead controller, a dead backplane, or a dead server.
- 03A real SAN. Storage as its own engineered system: a dedicated array with two controllers, its own drives, and its own network to reach it. The array has a life independent of any single compute node, which changes what a bad day looks like.
You cannot tell these tiers apart from the copy, because the copy does not say. The same five patterns appear across the category, paraphrased here because they are ubiquitous norms rather than any one company's sin.
| WHAT THE PAGE SAYS | WHAT IT NEVER SAYS |
|---|---|
| "Ultra-fast NVMe SSD storage" | Whether the drives are local or on an array, consumer or enterprise, behind one controller or two. A speed adjective carries the whole sentence. |
| "Enterprise SSDs" | Enterprise here describes the drive's marketing tier. It does not claim an enterprise array, because there usually isn't one behind it. |
| "RAID-10 protected storage" | The RAID card sits inside the same server. A dead drive is survivable; the storage still dies with the box. |
| "Up to N IOPS" | A ceiling with no latency figure, no floor, and no measurement under load. Ceilings are what benchmarks produce on an idle system. |
| "Redundant storage" | What exactly is redundant. Usually it means RAID again. A dual-controller array with multipath is a different machine, and copy that has one says so. |
None of these sentences say the storage is a separate redundant array, all-flash, reached over two independent fabrics, that survives the death of any single server. That sentence costs real money to be able to print, so pages that can print it, do. When the copy stops at the adjective, the architecture usually stops there too. That is the tell.
A page says "enterprise SSD" and stops there. Which machine is behind the word?
Drives behind one RAID card
Consumer parts chosen by the gigabyte, inside the server that runs your workload. Survives a dead drive and nothing else the chassis has planned for it.
A dedicated array
Two controllers, its own drives, its own fabric. Copy that can name an array does, because there is one to name.
A SAN keeps your data alive when the server running your workload dies
The point of a SAN is separation of fate. Your disk is a LUN on a dedicated array, and the server running your workload is just the machine currently mounting it.
- A CONTROLLER DIES
- The array's second controller keeps serving. On a RAID card there is no second controller.
- A SWITCH, CABLE, OR PORT DIES
- The other fabric carries the load, active/active.
- THE SERVER DIES
- The LUN keeps serving, and the workload comes back on another host against the same storage. On local disk or a RAID card, this is the day you find out how the backups were doing.
Speed is the part the copy always leads with, so here is ours in the only phrasing we will put in print: our arrays serve at single-digit-millisecond or better.
We standardized on one storage platform and learned it cold
We standardized on HPE 3PAR all-flash over Fibre Channel: one platform, chosen and mastered. We deviate only when a job genuinely earns it. Standardization is what buys the depth behind every answer in this post. The general argument for opinionated hardware is in its own post.
Your own array runs with no meter on it
Rented block storage bills the disk by the gigabyte, again by the provisioned IOP, and again on the way out. It's your array; there's no per-IOP meter and no egress toll on your own data. The full economics of the disk meter, including what those cloud line items actually charge for and what a real peak costs on owned flash, is in the meter on your disk.
Five questions expose any host's storage in one email
Everything above compresses into five questions. They take one email to send, and they are very hard to answer impressively without a real SAN behind you. Here they are, with our answers on the record.
Is my data on a SAN, or on a RAID card inside my server?
This is the fate-sharing question. If the honest answer is a RAID card, everything else on the page is describing one server. Our answer: a SAN. Your data lives on a dedicated, dual-controller all-flash array. It does not sit on a RAID card inside your server.
Is the storage all-flash?
Vague copy hides mixed tiers, and mixed tiers mean your latency depends on which shelf your blocks landed on. Our answer: we standardized on HPE 3PAR all-flash over Fibre Channel.
Is there multipath over independent fabrics?
Redundant means little until you ask what is independent from what. Two paths through one switch share that switch's bad day. Our answer: yes. Every host reaches its storage over two completely independent Fibre Channel fabrics with separate switches, running active/active. A switch, a cable, or a port can fail and your LUN keeps serving on the other path.
Where is my data replicated, and at what RPO?
The weak answer is a backup schedule. A backup measures how much you can lose since last night; replication is about how much you can lose at all. Our answer: replicated to a second array, with the RPO designed per workload.
Who fixes the array when it misbehaves, you or a vendor queue?
This is the question that separates operators from tenants. Our answer: we do. We keep engineers on staff who can read the array's internals and diagnose it at firmware depth, so the people who answer your call understand the storage down to that level instead of waiting on the manufacturer.
Count the nouns. A real answer names an architecture: an array, a fabric, a controller count, an RPO. A weak answer returns adjectives and a benchmark. And if question five produces a pause, you have learned where the storage actually sits.
Straight answers
What is a SAN in web hosting?+
Storage run as its own engineered system instead of drives inside your server: a dedicated array with two controllers, its own drives, and its own network to reach it. Your disk is a LUN on that array, and the server running your workload is just the machine currently mounting it, so the storage has a life independent of any single server.
Isn't local NVMe faster than a SAN?+
On an idle benchmark, a local NVMe drive will post very fast numbers. The figure that matters in production is service time under real load and what happens when hardware fails. Our arrays serve at single-digit-millisecond or better, and the LUN keeps serving through a dead server, a dead switch, or a dead controller. Local disk wins the idle benchmark and shares every fate the box has.
My host says RAID-10. Is that enough?+
RAID answers exactly one failure, a dead drive. It does nothing for a dead controller, a dead backplane, or a dead server, and on a RAID card any of those takes your storage with it. RAID inside one box and a dual-controller array on independent fabrics are different machines that happen to share the word redundant.
How should I use the five questions?+
Send them to your current host before you send them to a new one, and grade the replies on nouns. A real answer names the array, the fabric, and the person who fixes it. If the reply is adjectives and a benchmark, you have learned what the word enterprise meant on the pricing page.
Ask us the five questions
A storage engineer answers them for your actual workload, on the record, the same way we answered them here.
Talk to an engineerYou’ve reached the end.
Put the reading to work
What are
you running?
Tell us about the workload, traffic and recovery requirements. Start with the server builder, or talk through the details with an engineer.
Written by the engineers who build and run the infrastructure it describes.Talk to an engineer