Demand on systems is never constant. It rises, falls, has predictable peaks and unexpected surprises. Understanding your system's load patterns is essential for sizing infrastructure, planning capacity, and avoiding surprises.
This article explores the most common load patterns, how to identify them, and how to use this knowledge for better decisions.
Knowing your load patterns is knowing the rhythm of your business.
Why Patterns Matter
Correct sizing
If you size for the average, you'll have problems at peaks. If you size for the peak, you waste resources most of the time.
Understanding patterns allows:
- Effective use of autoscaling
- Intelligent capacity planning
- Preparation for known events
Anomaly detection
When you know the normal pattern, deviations become visible:
- Traffic outside expected hours may indicate an attack
- Absence of expected peak may indicate a problem
- Gradual pattern change may indicate product change
Common Temporal Patterns
Daily pattern (diurnal)
The most common: low load at night, growing in the morning, peak during business hours or evening.
Load
│
│ ╭──────╮
│ ╱ ╲
│ ╱ ╲
│ ╱ ╲
│─────╱ ╲─────
└────────────────────────────
0h 6h 12h 18h 24h
Examples:
- B2B SaaS: peak during business hours (9am-6pm)
- E-commerce: peak in the evening (7pm-10pm)
- Streaming: peak in late afternoon and evening
Weekly pattern
Significant variation between days of the week.
Examples:
- B2B: Monday to Friday intense, weekend low
- E-commerce: Friday and weekend more intense
- Entertainment: weekend is peak
Monthly pattern
Some systems have peaks related to the monthly calendar.
Examples:
- Payroll: end and beginning of month
- Accounting: monthly closing
- Subscriptions: renewal on specific dates
Seasonal pattern
Variations throughout the year.
Examples:
- E-commerce: Black Friday, Christmas, Mother's Day
- Education: semester start, enrollment period
- Tourism: school holidays, long weekends
- Accounting: end of fiscal year
Event Patterns
Scheduled events
Predictable peaks caused by known actions:
- Marketing campaigns
- Product/feature launch
- Media appearances
- Webinars and live events
Characteristic: you know when it will happen
Unscheduled events
Unexpected peaks:
- Organic virality
- News about the company
- Competitor failure
- External events (elections, crises)
Characteristic: you don't know when it will happen
The "Hacker News effect"
Also called "slashdot effect" or "hug of death":
Load
│
│ │
│ │
│ │╲
│ │ ╲
│────│ ╲────────
└──────────────────
Viral
An extreme and rapid peak, followed by gradual decline. Unprepared systems collapse.
Growth Patterns
Linear growth
Load
│ ╱
│ ╱
│ ╱
│ ╱
│ ╱
└──────────────
Time
Constant and predictable increase. Easy to plan for.
Exponential growth
Load
│ │
│ ╱
│ ╱
│ ╱
│────╱
└──────────────
Time
Common in viral products or after product-market fit. Hard to keep up with.
Step growth
Load
│ ┌──
│ ┌───┘
│ ┌───┘
│──┘
└──────────────
Time
Discrete jumps caused by new large customers or features.
How to Identify Your Patterns
1. Collect historical data
You need at least:
- 4 weeks for daily/weekly patterns
- 3 months for monthly patterns
- 1-2 years for seasonal patterns
2. Visualize at multiple scales
Look at the same data at different granularities:
- By hour (last 24h)
- By day (last month)
- By week (last year)
3. Identify correlations
What causes variations? Correlate with:
- Times and days
- Marketing events
- Product releases
- External events
4. Document and share
Create a "load calendar" that the whole team knows.
Planning for Patterns
Baseline + headroom
- Identify your baseline (typical minimum load)
- Identify your regular peaks
- Maintain headroom to absorb variations
Planned capacity = Regular peak × 1.5 (safety margin)
Elastic scaling
Configure autoscaling based on:
- Metrics that reflect real demand
- Reaction time appropriate to the pattern
- Limits to avoid excessive costs
Event preparation
For known events (Black Friday, launches):
- Project expected load
- Test required capacity
- Pre-scale before the event
- Actively monitor during
- Scale down after
Common Pitfalls
1. Assuming peaks are anomalies
If it happens regularly, it's not an anomaly — it's your pattern. Size for it.
2. Ignoring the "new normal"
After a growth event, the baseline may have permanently changed. Re-evaluate.
3. Planning only for average
The average is useful for costs, but peaks are what break systems.
4. Not correlating with business
Technical teams often don't know about marketing campaigns. Integrate calendars.
Conclusion
Load patterns are your business's temporal signature. Knowing them allows:
- Intelligent infrastructure sizing
- Advance preparation for events
- Problem detection through anomalies
- Data-driven communication with business
Invest time in understanding and documenting your patterns. It's knowledge that pays continuous dividends.
Your system doesn't operate in a vacuum. It operates to the rhythm of your business and your users.