Skip to content

cost-forecast.md

Latest commit

 

History

History
80 lines (54 loc) · 2.59 KB

cost-forecast.md

File metadata and controls

80 lines (54 loc) · 2.59 KB
Raw

Cloud Spend Forecasting

This document describes the cost forecasting methodology used by GistPin, covering how projections are generated, the assumptions behind the model, and how budgets are managed.

1. Forecasting Methodology

The cost-forecast.py script uses AWS Cost Explorer API data to build spend projections.

Data Sources

  • AWS Cost Explorer — daily unblended cost and usage, grouped by service.
  • AWS Rightsizing Recommendations — EC2 instance right-sizing suggestions for savings detection.

Model

A linear regression over historical daily costs is used to project future spend. The slope of the regression line represents the average daily cost change, extrapolated over 30-, 60-, and 90-day horizons.

Formula:

y(t) = α + β · t

Where:

  • y(t) is the projected cost at time t
  • α (intercept) = ȳ - β · x̄
  • β (slope) = Σ((xᵢ - x̄)(yᵢ - ȳ)) / Σ((xᵢ - x̄)²)

2. Model Assumptions

Assumption Rationale
Spend follows a linear trend Suitable for steady-state workloads; does not account for step changes (new deployments, traffic spikes)
Historical data is representative 90-day lookback captures seasonal patterns
Resource counts remain stable The model does not auto-detect scaling events
USD constant dollars No inflation or pricing changes factored in

Limitations

  • Linear regression under-forecasts during rapid growth phases (e.g. after a product launch).
  • No seasonality modelling — weekly/monthly patterns are averaged out.
  • No anomaly scrubbing — one-off charges (e.g. reserved instance purchases) distort the trend.

3. Budget Management

Budget Thresholds

Level Threshold Action
Info < 85% of budget Monitor
Warning 85–100% of budget Review cost-optimisation.sh output
Critical > 100% of budget Immediate spend review, restrict non-essential resources

Prometheus Alerts

Alerts defined in budget-alerts.yml fire when:

  • Projected spend exceeds budget
  • Forecast growth rate exceeds 20%
  • Resource cost growth exceeds 50%
  • Rightsizing savings exceed $100/month

Recommended Cadence

  • Run cost-forecast.py daily via cron or scheduled CI workflow.
  • Review budget-alert dashboard weekly.
  • Conduct a full cost review monthly.

4. Usage

# Text output (default)
python3 infrastructure/scripts/cost-forecast.py

# JSON output for downstream processing
python3 infrastructure/scripts/cost-forecast.py --output json

# Custom budget threshold
python3 infrastructure/scripts/cost-forecast.py --budget 10000