This is an opinion of the author – every environment and budget is different, so these recommendations may not be appropriate for you.
In my IT environment, we desire a certain minimum standard for all server hardware which I will try to outline here.
- Redundant power supplies – if one power supply fails or is accidentally unplugged, the other will power the server. This is not a feature you want to skimp on – it’s worth the money to have the server running while you wait for the replacement power supply to arrive when one fails.
- All hard-drives must be hot-swappable, NO internal direct-cabled hard drives allowed. This is to allow easy identification of a failed disk by the error light on the front of the carrier (drive sled). The failed disk can then be replaced with a good part while the system is running.
- All storage controllers must have a battery or flash-backed write cache. This enhances write performance on RAID volumes, and helps avoid data corruption when the system loses power in the middle of a write operation. The battery or flash write caches are often the distinguishing factor between a good RAID card and a bad one (even from the same vendor).
- RAID configuration – all RAID volumes (LUN’s) must be configured to allow the system to operate after detecting the failure of one drive. Acceptable RAID levels include 1, 1+0 “10”, 5, 5+0 “50”, and similar redundant configurations. Plain RAID 0 is not acceptable for any production data that you want to be available if (when) a disk fails. If you think disks don’t fail – think again, the more disks you have – the more disk failures you will see (basic probability).
- All disks that must perform well during system boot and disk write operations MUST be 10k RPM or faster SAS (Serial Attached SCSI). Fast high-performance disk operations are often referred to as “online” or “transactional” – a database is an excellent example of software that is both online and transactional – write performance matters! SATA disks are only acceptable for slow operations such as data archive, backup storage, etc (these slow operations are often referred to as “midline” or “nearline”). It is rare to find SATA disks that perform well enough for a production server that needs good disk write performance. Save yourself the time and trouble – use SAS storage for good performance online transactional systems.
- If you’re using a SAN or NAS for system storage, make sure the connection is FASTER than 1G. This means you must use Fibre Channel, 10G ethernet, or similar high-speed storage connectivity. This is not a place to save money, storage network bottlenecks on 1G ethernet are the most common system bottleneck for low-end or small-business systems. Don’t make the same mistake, Use FAST Storage Connections!
- Server hardware must be rack-mountable, and server rails should be the quick no-tool style. Mount all servers in standard 4-post square-hole racks. Power supplies should be routed to different UPS / power sources whenever possible (allowing the system to function during maintenance or failure of a single UPS).
- Minimum three year hardware warranty with next business day replacement of failed components. Opt for defective media retention (DMR) if the option is available – this allows you to keep and personally destroy any failed hard drives. Your production systems are likely to have proprietary or sensitive data which should be securely destroyed when a drive fails – pay the extra money to keep those failed drives from leaving your sight while you find a way to shred that old data (physically or digitally destroying all disk contents).