Skip to content

New: How We Design Technology - our seven-layer architecture model, explained in full. See the architecture

general

Performance, Availability, Cost: You Can't Max All Three

10/25/2026

The CFO wants to know why the infrastructure proposal costs more than last year's. The operations director wants the systems to stop slowing down every afternoon. The IT manager wants a second site so an outage won't interrupt work. All three requests are reasonable. Put together, they can produce an expensive project nobody has actually agreed to fund.

A good IT infrastructure strategy needs a shared language for those tradeoffs. At CTP, we use three words: performance, availability, and cost. The work is to decide what the business actually needs, understand what each option costs, and write down the reasoning so it can be revisited later. That rarely means buying the cheapest product or chasing the highest possible uptime.

At a glance

  • Define performance by real user workflows, availability by the disruption the business can tolerate, and cost as a multi-year total.
  • Different systems deserve different targets. Maxing out every characteristic for every workload is usually unnecessary and expensive.
  • Before committing to a vendor, document the options, assumptions, approval, proof of success, and the next review date.

Three words that mean different things to different people

Performance is whether people can get their work done without unreasonable delay. A loan officer opening a large document, a forensic analyst pulling instrument data, and a field employee joining a video call each run into different kinds of latency. The slow part might be storage, networking, authentication, the application itself, or a busy external service.

Availability is the ability to keep the business running, or bring it back, through a disruption. It includes redundancy, backup, disaster recovery, support, connectivity, tested fallback procedures, and people who know what to do. A redundant server won't help much if a single identity provider outage keeps every employee from signing in.

Cost is the total burden on the business over time: licenses, storage, network services, implementation, monitoring, backup, security, support, training, renewal risk, and the price of downtime. Sometimes paying more for one control reduces risk or effort somewhere else. Sometimes the sensible answer is to keep a simpler system.

Improving one of the three doesn't always hurt another. Better architecture can improve performance, resilience, and cost at the same time by removing waste. There is still no setup that maximizes all three for every scenario, so the tradeoffs have to be made deliberately.

Start your IT infrastructure strategy with the work that can't stop

Before anyone talks about products, ask each department to describe its critical workflows and what happens when they're delayed. A community bank may put teller systems and customer-facing services first. A forensic lab may prioritize LIMS availability and evidence integrity while accepting a longer recovery window for administrative email. A growing company may rank order processing and payroll well above a marketing file share.

For each workflow, set a response time or throughput target where it matters, an acceptable amount of downtime, and how much work can be lost or re-entered after recovery.

Recovery time objective (RTO) and recovery point objective (RPO) are useful shorthand, but they need translating. An RTO of four hours is a target for restoration, and it becomes a commitment only when the architecture and regular exercises back it up. An RPO of one hour means a recovery could lose, or require you to reconcile, up to about an hour of changes in the scenario you planned for.

Different systems earn different levels of investment. It's reasonable to design a critical transaction service for fast failover and accept slower recovery for an archive nobody opens often. A blanket "zero downtime" requirement applied to every workload usually means no one has ranked priorities yet.

A hypothetical decision: replace an aging file server

Imagine a 75-person business with an aging file server shared by finance, engineering, and administration. Staff complain about slow remote access. IT proposes moving everything to a single cloud file platform. The CFO wants to avoid another large hardware purchase. The operations director worries about losing access when the internet goes down.

The better first move is to separate the workloads and let each one land where it fits. General collaboration files might suit SharePoint or another SaaS service. A latency-sensitive engineering application may run better on right-sized local infrastructure or a specialized hosted service. A critical data set might need a tested backup and a documented offline procedure. For remote users, network connectivity and identity access could matter more than where the files are stored.

  • Retain and refresh local storage
    • Potential performance implication: Strong local access for suitable workloads
    • Availability consideration: Local failure and remote-access dependencies need design
    • Cost consideration: Upfront capital, plus lifecycle and support expense
  • Move selected work to SaaS
    • Potential performance implication: Convenient remote collaboration where the fit is good
    • Availability consideration: Internet, identity, vendor, and backup dependencies
    • Cost consideration: Subscription, migration, and governance costs
  • Deliberate hybrid design
    • Potential performance implication: Each workload in a suitable location
    • Availability consideration: More dependencies to document, though clear boundaries can improve resilience
    • Cost consideration: Requires integration work; can reduce overspending on mismatched workloads

Treat these as decision factors, since none of them guarantees performance. A proof of concept should test the actual application, file sizes, locations, identity policies, and network conditions before anyone approves the architecture.

Use the seven layers to expose hidden dependencies

We evaluate seven layers separately: identity and device management, collaboration, security monitoring, backup and portability, business applications, physical infrastructure, and connectivity. Separating them makes sure a decision at one layer doesn't quietly force decisions at all the others. It rarely means buying seven separate products.

Moving an application to SaaS, for example, might cut on-premises maintenance while increasing your dependence on identity, the provider's availability, internet links, and possibly a new backup model. A dedicated network service or SD-WAN may improve connectivity, with economics that depend on actual traffic and service commitments. A local Nutanix or Hyper-V system can make sense for one workload, while another organization needs no new local hardware at all.

The real value of the model is the conversation it starts. It lets leadership ask, "If this component fails or gets too expensive, what changes, what stays, and how do we recover?" That question stays useful long after any vendor comparison has gone stale.

A decision record worth keeping

Every significant architecture decision should be short enough for a director to understand and specific enough for the next IT leader to verify. Record the business requirement, the systems affected, the alternatives considered, target metrics, risks and dependencies, total cost over an agreed period, validation results, the owner, the approval, and the conditions that would trigger a second look.

A good entry might read: "We chose cloud collaboration for general documents because remote access and versioning matter more than local latency. We kept the engineering application on a small local platform pending compatibility testing. Recovery tests must meet the agreed targets, and we'll revisit the decision at the next network contract renewal." That tells a future team what was chosen and why.

Where uncertainty is high, approve a measured pilot. Collect real response times, failure behavior, support effort, and recurring cost. If the pilot misses its acceptance criteria, stopping should be an acceptable outcome. Approval gates only mean something when a project is allowed to fail them.

Cost needs a two-year view

The implementation quote is usually the easiest number to get and one of the least useful on its own. Ask for subscription growth assumptions, likely renewal events, ongoing support, backup, monitoring, network changes, and internal labor. Include transition or exit costs where they could be significant. Don't assume last year's pricing will hold.

For a small business, this can be a simple forecast with a base case and one stress case. For a regulated institution or lab, it may also need a documented risk decision about continuity and evidence obligations. If an additional service removes a meaningful single point of failure, name it and explain how you'll prove it. If the service only adds another dashboard, challenge the expense.

How to revisit the balance without starting over

Architecture decisions have a shelf life. Growth, staffing, new applications, new regulations, vendor consolidation, and price changes all shift the right balance. Review major decisions at contract events and on a regular governance cadence. Ask whether the original requirements still apply, whether the tests still pass, and whether a simpler or more resilient option is now available.

This is where vendor independence pays off. Rebalancing performance, availability, and cost is far easier when identity, collaboration, backup, monitoring, infrastructure, and connectivity are well understood and, as far as practical, replaceable. That gives you the freedom to make the next business decision on your own terms, without signing up for a migration that never ends.

What to bring to the next planning meeting

Bring your three most important business workflows, the interruptions each one can tolerate, the user complaints you can measure, current technology spending, and the date of your next contract renewal. With those in hand, you can have a serious conversation about tradeoffs before anyone picks a product.

The best IT infrastructure strategy is one you can explain, test, afford, and change when the business changes. That's how we design at CTP, with the reasoning behind every significant choice written down. Read more about how we design technology, or start a conversation grounded in your own environment.

Sources and further reading

Ask about something important for you.

Curious about our take on an industry or technology topic? Let us know.

Contact Us