IT Capacity Management for Cloud and Web Applications

What is Capacity Management?

Capacity management is the ongoing practice of making sure your IT systems have enough capacity to meet current and future demand, at a cost that fits your organization’s budget.

It covers everything from observing how much capacity you’re using currently, forecasting how demand will change in the future, verifying what your systems can actually handle, and tuning or scaling proactively to accommodate predicted demand.

The term carries enterprise connotations: ITIL 4 formalizes it as the capacity and performance management practice, complete with sub-processes and roles. You’ll also see the same practice called IT capacity management, infrastructure capacity management, or cloud capacity management, depending on which layer of the stack a team is responsible for.

This guide is the practical version for smaller teams as well as enterprise teams responsible for websites, web applications, and APIs.

After all, the work is essentially the same, whether you call it a formal capacity management process or simply “making sure we don’t crash.”

Capacity Management versus Capacity Planning

The terms “capacity management” and “capacity planning” overlap enough that they’re worth reviewing.

Capacity planning is the forward-looking piece: estimating future demand and deciding what infrastructure and configuration you’ll need to meet it. It’s a project or process with a deliverable, often run before a launch, a busy season, or an upcoming budget cycle. Check out the companion guide to capacity planning for web applications that covers that process in depth.

Capacity management is the continuous practice that planning is a part of. It includes the planning aspect, but also the monitoring that feeds it information, the load tests that validate current reality, and the tuning that stretches it to accommodate future demand. In short, capacity planning answers “what will we need next year?” while capacity management makes sure capacity is understood and planned for on an ongoing basis.

The Capacity Management Cycle

In practice, capacity management involves four main activities.

Monitoring Current Usage and Performance

The first activity is monitoring and calculating current usage. Take account of where you stand currently. Track your traffic levels, response times, error rates, and resource utilization over time rather than only glancing at them during incidents.

Two numbers matter most when it comes to understanding current capacity: your peak-to-average ratio and your utilization trend.

Your peak-to-average ratio tells you how concentrated your traffic is. It can be computed on different time intervals, such as hourly, daily, or even weekly. For example, a site averaging 1,000 visits an hour that averages 4,000 in its busiest hour has an hourly peak-to-average ratio of 4. A high peak-to-average ratio means your usage fluctuates a lot between peak and non-peak times. Some sites are much busier at certain times of year than others: for instance, an ecommerce site that has big Black Friday sales might have a very high peak-to-average ratio between Black Friday and a random Tuesday in June.

Your utilization trend tells you the rate of change in how much of your current capacity you’re using. If utilization is still within capacity but trending sharply upwards, you’re going to need more capacity soon or you risk slowness and outages.

Assessing and computing these numbers requires real-world data gathering. Analytics, APM, and infrastructure metrics provide the actual traffic and resource utilization data. Running synthetic monitoring with tools like Loadster Site & API Monitoring adds an external history of availability and response times.

Analyzing Trends and Forecasting Demand

The second part of capacity management is turning historical observations into forward-looking expectations.

Your system’s growth trends set the baseline; seasonality and your release calendar introduce the possibility of spikes. Your marketing team knows about an upcoming campaign before your servers do, so capacity forecasting is partly about communication: planned launches, promotions, press releases, and partnerships can all convert into higher traffic expectations.

The output of the forecasting process doesn’t need to be especially complicated. A finding like “We expect peak concurrency to double on Black Friday” is a useful forecast.

Validating Capacity with Load Testing

Here’s the step that takes much of the guesswork out of capacity management. Historical monitoring tells you how much capacity you’ve been using, but only testing tells you how much you actually have. The difference between those two numbers is your current capacity headroom.

Teams who do load testing as part of capacity management are routinely surprised by the results, in both directions.

The mechanics of load testing are discussed in how many concurrent users can your website handle, but in short, ramp realistic simulated traffic past your projected peak to find the level where performance stops meeting your standards.

If performance is fine when simulating current peak levels of load, but degrades substantially at double that level, it’s a clear indication that you’re already above 50% capacity and need to add headroom to accommodate further growth.

Make a point to rerun load tests after significant releases and architecture changes, because your capacity shifts silently as the system changes. A new feature that introduces one expensive query might lower your capacity ceiling by half, raising the risk of crashes at the busiest possible time.

Tuning Applications to Extend Existing Capacity

Another activity is squeezing more performance from what you already have, which is usually cheaper than buying more capacity. Caching, query optimization, connection pool sizing, and moving slow work to background queues all raise the ceiling without raising the infrastructure bill. Our performance tuning guide lists many of the easy wins that are worth pursuing.

Tuning and validation form a natural loop: test, fix the bottleneck, retest, and record the capacity number. When tuning stops paying off, that’s your signal that the next increment of capacity has to come from additional infrastructure. This feeds back into capacity planning.

Capacity Management in the Cloud

Autoscaling changed capacity management less than advertised. It was framed as an easy solution, and it solved one problem: stateless compute can now grow with demand instead of being provisioned for the worst case with hard limits.

But most systems are only partly stateless, and the parts that are stateful don’t autoscale so well.

Traditional databases are the most obvious example of this: connection limits, IOPS, and replication lag don’t add themselves when traffic doubles.

The same goes for third-party API rate limits, license-capped software, and anything in your stack with a fixed resource pool.

Autoscaled web tiers may scale up sharply in response to increased load, but this only serves to deliver overwhelming traffic to these non-scaling components in your stack faster than ever! This is one reason why load tests so often reveal the capacity bottleneck lurking just one layer further down.

Moving to cloud platforms also changed the failure mode. On fixed infrastructure, a capacity mistake might quickly become an outage when demand exceeds capacity. With autoscaling, the failure mode might be a catastrophically high bill instead. Either way, it’s a capacity management problem.

Knowing what doubling in traffic costs, in terms of both performance and money, is the cloud era equivalent of knowing how many servers you need to rack up.

Capacity Management Best Practices

The habits that make an ongoing capacity management practice successful are fairly straightforward.

• Track trends, not snapshots. Utilization and response times over months inform when capacity is likely to run out, not a one-time glance at a dashboard.
• Know your peak-to-average ratio, and do all capacity math with the highest peaks in mind.
• Validate capacity with load testing, especially before your busy season and after major changes to your stack, rather than simply extrapolating from last year’s number.
• Keep a headroom target. Many teams aim for verified capacity of at least 2x the expected peak. Pick a ratio that matches your risk tolerance.
• Give your capacity model an owner. A shared spreadsheet nobody owns is easily ignored.
• Retest after every significant tuning or scaling change. Make sure the model reflects the system you actually run.

Tools for Capacity Management

No single product does capacity management for you. It requires inputs from multiple products. Monitoring and APM tools cover the historical measurements, load testing platforms cover the validation and tuning, and a spreadsheet might be all it takes for the forward planning calculations.

Loadster handles two of the three tool needs: load testing to measure your real capacity and synthetic monitoring to watch usage and performance trends between tests, with the same scripts working for both. If you’d like to put a real number on your current capacity, the free trial fuel is enough to run a few tests with no credit card required.