Justifying and Planning Server Selection
A good engineering choice is only half the job. Someone has to approve it, fund it, and live with it for years. This lesson turns a server decision into a case you can defend — with requirements, capacity numbers, cost, lifecycle, and documentation — and then closes the whole course.
Needs assessment
Before any specification, write down what the business actually needs. A needs assessment is a structured conversation that produces requirements you can test against. Capture three groups:
- Users and devices — how many people, how many endpoints, whether they are on-site or remote, and what each one does.
- Applications and services — every workload the platform must host (email, file sharing, a database, directory services), each with its own CPU, memory, IOPS, and licensing profile.
- Growth and constraints — expected headcount and data growth per year, plus fixed constraints such as a budget ceiling, a compliance rule, or an existing data centre.
The reason the distinction matters is accountability. A requirement can be tested — somebody can measure the delivered system against it and say yes or no. A wish cannot be, so it turns into an argument in the budget meeting. When you write a wish down, the useful next step is to convert it into a number: “make it fast” becomes “file opens must return under 200 ms at peak load.”
Capacity planning
Capacity planning is the exercise of turning requirements into sized resources. Two ideas do most of the work:
- Peak versus average. Average load flatters a server; peak load breaks it. Size for the busiest realistic moment — the Monday-morning file-open storm, the payroll run, the nightly backup — because that is when users notice.
- Headroom. Never plan to run at 100%. Healthy targets are commonly 50–70% sustained utilisation at peak, leaving room for spikes, growth, and the loss of one component. The margin is what turns an outage into a non-event.
Performance metrics
Capacity planning is not a one-off. To know whether a server is still right-sized, watch the same metrics in production:
| Metric | What it tells you | Warning sign |
|---|---|---|
| CPU utilisation | Compute pressure; sustained highs mean the workload outgrew the CPU | Sustained > 80% at peak |
| Memory utilisation | Whether the working set fits in RAM | High use plus disk paging |
| Storage I/O (IOPS, queue depth) | Whether storage can keep up with reads/writes | Growing queues and rising latency |
| Latency | Response time users actually feel | Steadily climbing under the same load |
| Network utilisation | Whether the NIC or uplink is saturated | Links pinned near 100% at peak |
TCO and ROI
Hardware is not the whole cost. Total Cost of Ownership (TCO) sums everything it takes to own and run the solution over its life:
- Hardware and licenses — servers, storage, networking, and the software licenses attached to them (often per core or per user).
- Power and cooling — electricity to run the gear and the air conditioning to remove its heat. This runs every day for years and is easy to underestimate.
- Facilities — rack space, physical security, and the space itself.
- Staffing and operations — the people who patch, monitor, back up, and troubleshoot it, plus the cost of downtime if it fails.
Return on Investment (ROI) asks a different question: does the money spent deliver a benefit worth more than the cost? ROI can be hard savings (fewer servers after consolidation, less power) or value such as reduced downtime, faster delivery, or a capability the business did not have. A sound case states both, honestly labelled.
Lifecycle, refresh, documentation and change management
Servers are not permanent. A typical lifecycle is three to five years: the period the vendor supports the hardware with firmware updates and spare parts under warranty. Plan the refresh before it arrives, not after:
- Track the age and support status of every server; out-of-warranty hardware is a risk with no safety net.
- Budget ahead so a refresh is a planned cost rather than an emergency purchase.
- Re-assess the requirement at refresh — the workload, the cloud options, and the price of hardware will all have changed since the last purchase.
- Retire securely — wipe or destroy storage media so data cannot be recovered from a decommissioned disk.
A server decision that is not documented cannot be defended, audited, or repeated. Two practices protect the investment:
- Documentation. Record the requirements, the chosen specification, the rationale for each major choice, the cost model, the network details (addresses, VLANs, management port), and the backup plan. When the server fails at 2 a.m., this is what rescues the on-call engineer.
- Change management. Changes to production infrastructure go through a controlled process: request, review/approval, a rollback plan, a maintenance window, and a record of what was done. It reduces outages caused by well-meant but untested changes and leaves an audit trail.
The consolidation decision tree
Three questions, asked in order, resolve most server decisions:
Used in order, the tree keeps you from jumping to the answer you like. Most workloads virtualize first, because one physical host running several virtual servers lowers the hardware count and makes snapshots and recovery easy. Only appliances, hard real-time latency, or licences tied to physical hardware justify staying physical. The second question separates the deployment models from Lesson 5, and the third separates building a platform from buying one.
Worked mini-case: 150 employees
The situation. A 150-employee company has outgrown shared folders and a consumer mail service. It needs a proper email platform and a central file platform. About 60 staff use email heavily; the file share will hold roughly 10 TB of documents and grow about 10% a year. Leadership wants a recommendation it can approve this quarter.
- Step 1 — needs assessment. Users: 150, with 60 heavy email users. Apps: email plus file sharing, plus room to grow. Constraints: sensitive internal documents, a modest IT team (one or two people), and a preference for predictable cost.
- Step 2 — capacity. Email for 150 users is a well-understood, largely outbound workload; the file share needs ~10 TB now and about 15 TB of headroom by year five. Peak concurrency is modest, so neither service is compute-heavy.
- Step 3 — apply the decision tree.
- Virtualize or physical? Virtualize. Both services are standard workloads that run well as virtual machines, which lowers hardware count and enables snapshots and easy recovery.
- On-prem or cloud? Split the answer. For an organisation this size, the email platform is a classic SaaS candidate: the provider runs the mail servers, spam filtering, and patching, and the cost is a predictable per-user subscription — a strong TCO result against building and staffing an on-prem mail system. The file platform, because the data is sensitive and access is local, leans on-premises or private/hybrid, where the data stays under the company’s control and access is low-latency.
- Build or buy? Buy the email (SaaS); build/own the file tier (a virtualized file server on existing or new hardware). Buying email keeps the small IT team focused on the business rather than on mail-server operations.
Step 4 — TCO. Email as SaaS turns a hardware-plus-license-plus-staffing project into a known recurring subscription. The file platform carries CAPEX for storage, plus power, cooling, and backup — but stays a modest, local cost. Both sides are compared over five years, not on purchase price alone.
Step 5 — lifecycle and documentation. Put both services on a five-year review. Document the file server’s spec, addresses, backup schedule, and the change process; document the SaaS tenancy, admin accounts, and data-location and contract terms. Revisit at refresh.
Capstone recap
Week 11 completes the infrastructure picture. Tick each capability:
You’ve completed the course
You started at Week 1 not knowing a PDU from a port. You have since learned how devices are addressed, how the layers fit together, how data is encapsulated, how IP and subnetting work, how the transport and application layers behave, and how to secure a network — and now, how servers are chosen, hosted, and justified. That is the full arc of CCNA 200-301, with the security fundamentals of CyberOps 200-201 woven through.
Two habits carry the knowledge forward: read the sources at the end of every lesson rather than trusting a summary, and practise in a lab or Packet Tracer until the commands and the reasoning are automatic. Theory that is never run is half-learned.
From here, keep a copy of the roadmap as your index, retake the quiz until the foundations are solid, and treat Week 11 as the template for every infrastructure decision you will make next. Congratulations — the course is complete, and you are exam-ready to keep building on it.
Review
Check what you have learned with these review questions.
Sources & references
- Mell, P. & Grance, T. NIST SP 800-145: The NIST Definition of Cloud Computing (2011). csrc.nist.gov.
- Microsoft Learn. Cloud economics and Total Cost of Ownership. learn.microsoft.com.
- Amazon Web Services. Total Cost of Ownership guidance. aws.amazon.com.
- Cisco Networking Academy. CCNA: Introduction to Networks (ITN), module resources on change management and infrastructure planning. netacad.com.
- Cisco. CCNA (200-301) Exam Topics. cisco.com.
Target exams: CCNA 200-301 v1.1 · CyberOps Associate 200-201. Last content verification: 2026-10-11.