Server selection starts with the workload: a database server and a file server have fundamentally different requirements. The six criteria that matter most are processor architecture, RAM capacity and ECC support, storage type and RAID configuration, network interface count, redundant power supplies, and on-site warranty response time.
1. Processor architecture and core count
A database workload is typically latency-sensitive and benefits from higher single-core performance. A virtualisation host needs core count more than clock speed — more cores means more VMs.
Intel Xeon and AMD EPYC are the two dominant enterprise server CPU families. Choosing between them is a workload question, not a brand question.
2. RAM and ECC support
ECC (Error-Correcting Code) RAM detects and corrects single-bit memory errors silently. All enterprise server platforms support ECC; consumer platforms do not. For a production database or virtualisation host, ECC is non-negotiable.
RAM sizing depends on the workload: a virtualisation host running ten VMs with 8 GB each needs at least 128 GB to leave headroom for the hypervisor and overhead.
3. Storage: NVMe, SAS or SATA — and RAID level
NVMe SSDs deliver the lowest latency for IOPS-intensive workloads such as databases. SAS HDDs remain cost-effective for large-capacity file or backup storage. SATA SSDs sit between the two.
RAID-10 (mirroring plus striping) gives both redundancy and read performance for database workloads. RAID-6 is common for large-capacity storage where two drive failures must be tolerated without data loss.
- RAID-1: simple mirroring, two drives, any workload
- RAID-10: performance + redundancy, minimum four drives
- RAID-5: space-efficient, write penalty, one drive failure tolerance
- RAID-6: two drive failure tolerance, larger arrays
4. Network interface count and speed
A server handling production traffic and backup simultaneously benefits from dedicated NICs for each traffic class. A minimum of two 1 GbE ports in a team (bond) is the baseline; 10 GbE is the standard for virtualisation hosts.
5. Redundant power supplies
A single PSU failure takes down a server with no redundant power. Hot-swap redundant PSUs (typically 1+1 configuration) allow a failed unit to be replaced without downtime.
For critical workloads, feed the two PSUs from separate PDUs on separate circuits where possible.
6. Warranty and on-site response
A next-business-day parts-only warranty means waiting a day for parts and then finding your own engineer. Next-business-day on-site warranty means a manufacturer-certified technician arrives with parts.
For production systems, on-site warranty with a 4-hour response (NBD on-site or better) is the minimum to specify. Check the warranty terms for your postcode — response times vary by location.
