Building a server is a very broad task - the list of stuff to check/consider will vary widely based on the server's intended role.
What follows are some guidelines. They are by no means a complete checklist. Your environment WILL vary, and local deviations, additions, and removals will be necessary.
Let's start at the beginning - When you're preparing to deploy a new server you will need to consider the hardware, particularly Capacity Planning issues.
- RAM - How much do you need?
- CPU - One, two, or more? Multi-Core (Almost certainly "yes" these days).
- Is your workload CPU-bound (faster core speed is better) or parallelized (more cores is better)?
- Is your workload Hyperthreading-Friendly, or should you turn off Hyperthreading?
- Disk - Capacity is the obvious thing.
- "Enterprise Features"
- Lights-Out Management, IPMI, on board KVM, etc.
- Single or Redundant power supply
- Hot-Swap or internal/cabled disks?
- Form Factor ("tower server", 1U, 2U, etc.)
- Power and Cooling Requirements
The obvious "Windows or *NIX" question aside, the same basic things need to be considered no matter what OS you run:
- Partitioning - Do you have a company-wide standard layout?
What layout makes sense for this system?
- Network Configuration - IP, subnet, gateway, DNS.
- Firewall rules (Local? Network core?)
- User Management
(Local users?, Centralized in AD/LDAP/NIS?)
- What standard local accounts/passwords do you add?
Common Configuration & other tasks
No matter the role, most servers will need:
- Backup - Agents, schedule, etc.
- Monitoring - At least ping and/or SNMP. And any key services.
This is where things will diverge considerably: Are you building a DNS server? A mail server? An Active Directory Domain Controller? Each will have its own (role- and site-specific) checklist.
I am not even going to pretend to offer guidance here. There's just too much variability.
Remember to record the existence of the sever (serial number, location/rack/U, host name, role, OS, etc.)
Also remember to LABEL the physical server. Many production-critical systems have met their demise because there was no label on the physical machine, and since nobody knew what it did it got unplugged.