Interactive design lab

Load Predictor

What is load prediction?

Load prediction turns recent demand, growth, recurring peaks, and known events into a planning range. The range matters more than a precise-looking point because traffic changes and model error can consume capacity before a scaling response is ready.

Step 1

Model

Fit the historical signal with a baseline, trend, volatility, and measured backtest error. Then shape seasonality and events, set a tested serving envelope, and choose how much capacity to reserve.

Step 2

Observe

Compare expected and challenged demand with usable capacity. Use the first saturation day, peak shortfall, requests at risk, and required fleet size to define an alert and scaling action.

Step 3

Challenge

Start with the expected range, then inject a launch spike, trend break, flash crowd, regional failover, and forecast error. Notice which failures increase demand and which remove capacity.

Demand forecast workbench

Predict the load, then challenge the plan

Fit a transparent trend to recent traffic, shape the next peak, and test whether usable capacity survives uncertainty and abrupt demand changes.

Loop 1

Fit the historical signal

The baseline, trend, volatility, and backtest error move the centerline and planning band.

Current baselineRPSHistory windowweeks

Observed weekly trend4%A fitted teaching trend, not a production statistical estimate.Observed volatility18%Controls visible variation in the historical signal.Backtest error14%Use an error measured on held-out history when operating a real forecast.

Loop 2

Shape seasonality and events

Change when and how demand peaks without changing the fitted historical baseline.

Forecast horizon30 daysLonger horizons widen the heuristic planning band.Weekly seasonality35%Adds a repeating weekday demand wave.Weekend adjustment-20%Use negative values for business traffic and positive values for consumer peaks.

Planned eventdayEvent liftmultiplier

Capacity loop

Set the serving envelope

Measured instance throughput, fleet size, and reserve determine the green capacity line.

Tested throughputRPS / instanceLive instancesinstances

Capacity reserve25%Reserved capacity is excluded from the usable serving line.

Forecast

Expected range: demand versus usable capacity

Demand follows the fitted trend, weekly shape, and planned event assumptions.

Capacity holds

Expected peak

11.3K RPS

Day 11 on the centerline

Scenario peak

11.3K RPS

Day 11 under baseline

Usable capacity

13.5K RPS

18K raw before reserve and failures

Peak headroom

+2.2K RPS

19% relative to scenario demand

Observed historyExpected baselineHeuristic uncertaintyChallenge demandUsable capacity

This teaching model compounds a transparent trend and weekly shape. The +/-20% band combines backtest error, observed volatility, and horizon drift. It is a moderate planning band, not a statistical confidence interval or production forecast.

Challenge mode

Break the healthy forecast

Each scenario changes demand or surviving capacity. The chart and consequences update together.

Consequence

Reserve absorbs the scenario

Capacity remains above the scenario peak, but the remaining reserve still determines response safety.

Scenario demand11.3K RPS

Usable capacity13.5K RPS

Peak shortfall

0 RPS

No modeled miss at peak

Requests at risk

15-minute peak exposure

Decision

Translate the forecast into an operating action

The useful output is a trigger and response, not a precise-looking point estimate.

First saturation: Not in horizon
Instances required: 11
Current provision: 12
Planning band: +/-20% (moderate)

Keep the current fleet, but alert before demand consumes the remaining 2.2K RPS. Refit the trend when measured error leaves the displayed band.

No quiz questions available

Could not load questions file