---

name: capacity-management description: > Plan IT capacity with demand forecasting, resource utilization analysis, scaling triggers, performance baselines, and capacity reports for proactive infrastructure management. TRIGGER when: user says /capacity-management, "capacity planning", "capacity management", "resource utilization", "demand forecasting", "scaling plan", "infrastructure capacity", or "will we run out of capacity". argument-hint: "[system_name or resource_type]" user-invocable: true

IT Capacity Management

You are an expert IT capacity management practitioner. When the user asks you to plan or manage IT capacity, follow this structured process to ensure infrastructure meets current and future demand while optimizing cost and performance.

Step 1: Current Capacity Assessment

Establish a baseline of current resource utilization:

Resource Type	Metrics to Collect	Source	Collection Method
Compute (CPU)	Utilization %, core count, clock speed	Servers, VMs, containers	Monitoring agent (Prometheus, Datadog, CloudWatch)
Memory (RAM)	Utilization %, allocated vs used, swap usage	Servers, VMs, containers	Monitoring agent
Storage	Used capacity, IOPS, throughput, latency	SAN, NAS, cloud storage, local disk	Storage management tool, cloud metrics
Network	Bandwidth utilization, packet loss, latency	Switches, routers, load balancers	SNMP, NetFlow, cloud VPC metrics
Database	Connection count, query throughput, buffer pool hit ratio	Database servers	Database monitoring (pganalyze, RDS metrics)
Application	Request rate, response time, error rate, queue depth	Application servers	APM tool (New Relic, Datadog APM)
Cloud resources	Instance count, reserved vs on-demand, spot utilization	Cloud provider	Cloud cost management tool (CloudHealth, Spot.io)

Utilization Summary Template

Resource	Current Capacity	Peak Utilization	Average Utilization	Available Headroom	Status
CPU — Prod cluster	128 cores	82%	55%	18% at peak	Warning
RAM — Prod cluster	512 GB	71%	48%	29% at peak	OK
Storage — Primary DB	2 TB	78%	78% (growing)	22%	Warning
Network — WAN link	1 Gbps	45%	22%	55% at peak	OK
DB connections	200 max	168 (84%)	92 (46%)	16% at peak	Warning

Health Status Thresholds

Status	Utilization Range	Action
OK (Green)	0-60%	Monitor normally
Warning (Yellow)	60-80%	Plan capacity increase, monitor closely
Critical (Red)	80-90%	Immediate scaling needed, alert stakeholders
Emergency (Black)	90-100%	Service impact imminent or occurring, emergency action

Step 2: Demand Forecasting

Project future capacity requirements:

Forecasting Methods

Method	Data Required	Best For	Accuracy
Linear trend	6-12 months historical utilization	Steady, predictable growth	Medium
Seasonal decomposition	12-24 months historical with seasonal patterns	Business with cyclical demand	High
Business driver correlation	Utilization + business metric (users, revenue, transactions)	Growth tied to business KPIs	High
Capacity modeling	Application architecture, resource-per-unit ratios	New systems, major changes	Medium
Scenario planning	Business plans, product roadmap	Strategic planning, uncertainty	Variable

Demand Forecast Template

Resource	Current	3-Month Forecast	6-Month Forecast	12-Month Forecast	Confidence
CPU cores	128	140 (+9%)	155 (+21%)	180 (+41%)	High
RAM (GB)	512	550 (+7%)	600 (+17%)	700 (+37%)	High
Storage (TB)	2.0	2.3 (+15%)	2.7 (+35%)	3.5 (+75%)	Medium
DB connections	200	220 (+10%)	250 (+25%)	300 (+50%)	Medium

Growth Driver Analysis

Business Driver	Current Value	Growth Rate	Capacity Impact
Registered users	500,000	+8% monthly	+2 cores, +8 GB RAM per 50K users
Daily transactions	1.2M	+5% monthly	+50 IOPS per 100K transactions
Data ingestion	50 GB/day	+10% monthly	+1.5 TB storage per quarter
API calls	10M/day	+15% monthly	+4 cores per 5M daily calls
Concurrent users	25,000 peak	+6% monthly	+500 DB connections per 10K concurrent

Step 3: Performance Baselines

Establish and monitor performance baselines:

Baseline Metrics

Metric	Baseline Value	Acceptable Range	Alert Threshold	Measurement Period
API response time (p50)	120 ms	80-160 ms	> 200 ms	Last 30 days
API response time (p99)	450 ms	300-600 ms	> 800 ms	Last 30 days
Database query time (avg)	25 ms	15-40 ms	> 60 ms	Last 30 days
Page load time	2.1 s	1.5-3.0 s	> 4.0 s	Last 30 days
Error rate	0.1%	0-0.3%	> 0.5%	Last 7 days
Throughput (requests/sec)	1,200	800-2,000	< 600 or > 2,500	Last 7 days

Baseline Deviation Detection

Deviation Type	Detection Method	Response
Gradual degradation	Week-over-week trend analysis	Investigate root cause, plan capacity
Sudden spike	Real-time alerting on threshold breach	Incident response, auto-scaling
Periodic pattern	Seasonal decomposition, day-of-week analysis	Adjust baselines, pre-scale for known peaks
Post-change regression	Before/after comparison around change windows	Rollback or optimize, update baseline

Step 4: Scaling Triggers and Strategies

Define when and how to scale:

Scaling Trigger Matrix

Resource	Scale-Up Trigger	Scale-Down Trigger	Method	Lead Time
CPU	Sustained > 75% for 15 min	Sustained < 30% for 1 hour	Horizontal (add instances)	5 min (cloud)
Memory	Sustained > 80% for 10 min	Sustained < 40% for 1 hour	Vertical (upgrade instance)	15 min (cloud)
Storage	> 75% capacity	N/A (rarely shrink)	Add volume, expand disk	30 min (cloud)
Database	Connections > 80% max	Connections < 30% max	Read replicas, connection pooling	1-4 hours
Network	Bandwidth > 70% for 5 min	Bandwidth < 20% for 1 hour	Upgrade link, add CDN	Hours-days

Scaling Strategies

Strategy	Description	Best For	Considerations
Vertical scaling (scale up)	Increase resource on existing instance	Databases, monoliths	Downtime during resize, upper limits
Horizontal scaling (scale out)	Add more instances behind load balancer	Stateless services, web tier	Requires load balancing, session management
Auto-scaling	Automated horizontal scaling based on metrics	Cloud-native, variable load	Set min/max limits, avoid thrashing
Pre-scaling	Proactively scale before known demand spikes	Events, sales, seasonal peaks	Requires advance knowledge of demand
Caching	Reduce load on origin systems	Read-heavy workloads	Cache invalidation complexity
Queue-based buffering	Absorb burst traffic, process asynchronously	Write-heavy, batch processing	Adds latency, queue monitoring needed

Auto-Scaling Configuration Template

Resource: [target group / service]
Metric: [CPU utilization / request count / custom metric]
Scale-out: When [metric] > [threshold] for [duration], add [count] instances
Scale-in: When [metric] < [threshold] for [duration], remove [count] instances
Minimum instances: [floor]
Maximum instances: [ceiling]
Cooldown period: [seconds between scaling actions]
Health check: [endpoint and expected response]

Step 5: Capacity Reports

Produce regular capacity reports for stakeholders:

Monthly Capacity Report Template

Section	Content
Executive summary	Key findings, risk areas, recommended actions
Resource utilization summary	Table of all resources with current/peak/average utilization
Trend analysis	Utilization trends with forecast overlay
Capacity risks	Resources approaching thresholds with timeline to breach
Cost analysis	Current spend, projected spend, optimization opportunities
Scaling events	Auto-scale and manual scale events in the period
Performance baseline status	Any baseline deviations and their causes
Recommendations	Prioritized list of capacity actions

Capacity Risk Register

Risk ID	Resource	Current Util.	Projected Breach	Business Impact	Mitigation	Owner	Cost
CR-001	Primary DB storage	78%	6 weeks	Service outage if full	Expand to 4 TB, archive old data	DBA Team	$500/month
CR-002	Prod CPU cluster	82% peak	4 weeks	Degraded response times	Add 4 nodes to cluster	Platform Team	$2,400/month
CR-003	WAN bandwidth	45% peak	6 months	Slow remote access	Upgrade link to 2 Gbps	Network Team	$1,200/month

Step 6: Optimization and Cost Management

Balance capacity with cost efficiency:

Optimization Opportunities

Opportunity	Detection Method	Potential Savings
Right-sizing over-provisioned instances	Instance utilization < 30% consistently	30-60% per instance
Reserved instance purchasing	Stable workloads running 24/7	30-70% vs on-demand
Spot/preemptible instances	Fault-tolerant, interruptible workloads	60-90% vs on-demand
Storage tiering	Infrequently accessed data on cold storage	50-80% on storage costs
Decommission unused resources	Resources with zero utilization for 30+ days	100% of unused resource cost
Scheduling non-production	Dev/test environments off during non-business hours	40-60% of non-prod costs
Application optimization	Code profiling reveals inefficient resource use	Variable — often 20-50%

Cost-Capacity Trade-off Analysis

Option	Capacity Provided	Monthly Cost	Cost per Unit	Lead Time	Flexibility
On-demand cloud	Variable	$$	High	Minutes	Maximum
Reserved instances	Fixed	$	Medium	Committed	Low
Spot instances	Variable	$	Low	Minutes	High risk
On-premises	Fixed	$ (amortized)	Low (at scale)	Weeks-months	Minimal
Hybrid	Mixed	$	Medium	Variable	Moderate

Output Format

Present the capacity management deliverable as:

1. Capacity Summary (current state, key risks, headline recommendations)
2. Resource Utilization Report (all resources with current/peak/average and status)
3. Demand Forecast (3/6/12-month projections with confidence levels)
4. Performance Baselines (metrics, acceptable ranges, current deviations)
5. Scaling Plan (triggers, strategies, auto-scaling configurations)
6. Capacity Risk Register (risks ranked by urgency with mitigations)
7. Cost Optimization Analysis (opportunities, savings estimates, trade-offs)
8. Capacity Roadmap (quarterly plan for capacity investments and optimizations)

Quality Checklist

Before delivering the capacity plan, verify:

• All critical resources are inventoried with current utilization
• Performance baselines are established with at least 30 days of data
• Demand forecasts use appropriate methods with stated confidence levels
• Scaling triggers are defined with specific thresholds and response actions
• Cost implications are quantified for all capacity recommendations
• Capacity risks have owners, timelines, and mitigation plans
• Report cadence and distribution are defined
• Optimization opportunities are identified and prioritized
• Plan accounts for known business events (launches, campaigns, seasonal peaks)

Edge Cases

• Sudden viral growth: Pre-define an emergency scaling runbook with pre-approved budget; use auto-scaling with generous maximums; have a CDN and queue-based buffering ready
• Multi-cloud environments: Normalize metrics across providers; use a unified monitoring platform; consider cross-cloud burst capacity
• Legacy on-premises systems: Plan longer lead times for hardware procurement (8-16 weeks); maintain a buffer of spare capacity; evaluate cloud migration for elastic workloads
• Shared infrastructure (multi-tenant): Implement resource quotas per tenant; monitor noisy-neighbor effects; plan capacity based on aggregate demand, not individual tenants
• Seasonal businesses (e.g., retail): Forecast based on prior year peaks with growth factor; pre-scale 2 weeks before peak; use reserved capacity for baseline + on-demand for burst
• Data growth outpacing compute growth: Implement data lifecycle policies (archive, purge, compress); evaluate storage tiering; consider data warehouse vs operational database separation