What makes a server an enterprise server

The box that looks like a bargain

A common budget offer is built from desktop parts in a rack case: a consumer motherboard, one power supply, one large SATA disk for data, a small consumer M.2 drive for the operating system, memory without error correction, and no way to reach the machine when it stops answering. Under normal conditions it runs. The difference between that machine and an enterprise server is invisible until the first component fails, and then it is the whole difference.

Memory that corrects itself

Enterprise servers use ECC memory, which detects and corrects single-bit errors and reports them. On a machine without ECC a flipped bit silently corrupts whatever was in that location: a database page, a file being written, a running process. Our servers log corrected errors, and a module that starts correcting too often is replaced before it fails. A desktop board has no way to know it is happening.

A write cache that survives a power cut

Database and file system performance depends on a RAID controller with a write-back cache. The enterprise version protects that cache with a battery or a flash module, so writes that the operating system believes are safe survive a power loss. A consumer SSD behind a basic or absent controller acknowledges writes it has not yet committed; a power cut loses them, and a database can come back inconsistent. The same applies to the drives themselves: enterprise SSDs carry power-loss protection capacitors and are rated by the terabytes they can write over their life; consumer M.2 drives are rated for laptops.

Two power supplies on two feeds

Every server we rack has two power supplies, each on a different feed from a different distribution path. A failed supply, a tripped breaker or maintenance on one feed changes nothing. A single-supply machine on one feed stops, and stays stopped until someone is physically in front of it.

Parts that swap while it runs

Drives, fans and power supplies in an enterprise chassis are hot-swappable: the failed part comes out and the replacement goes in with the server running and the operating system unaware. In a budget box, replacing a drive means powering down, opening the case and, usually, a reinstall.

A way in when the operating system is gone

Enterprise servers carry a management controller, iLO on HPE and iDRAC on Dell, on its own network port: a console that works when the operating system does not, power control, hardware health, and the ability to reinstall from a remote image. We use it over a management network separate from customer traffic. Without it, a kernel panic at three in the morning is a visit to the datacenter.

Firmware with a life

A server platform receives firmware for years: BIOS, controller, drive and network firmware with security fixes and compatibility updates, and a published end of support. The vendor keeps spare parts for that life, and so do we, on the shelf in the same building. A consumer board gets a handful of updates and then nothing.

What the two machines cost on the day it matters

On the day a drive fails, the enterprise server rebuilds its array while serving traffic and the customer learns about it from a ticket we opened. The budget box is down until a human arrives with a part, and if the write cache was unprotected, the restore starts from the last backup. The price difference between the two machines is smaller than one afternoon of that.

What we rack

Our dedicated servers are current HPE ProLiant and Dell PowerEdge machines. The example configuration on our dedicated server page is an HPE ProLiant DL380 Gen10 with two Intel Xeon Gold 6248 processors, 192 GB of ECC memory and a 2.4 TB all-flash volume on HPE 3PAR storage over dual-path Fibre Channel, with two power supplies, hot-swap everything and iLO on our management network, from EUR 500 a month, fully managed. See Dedicated servers and bare metal servers.