Up-scaling

Effective up-scaling is crucial for maintaining application performance while controlling costs. This guide explains how Stackbooster.io's intelligent up-scaling works to ensure your Kubernetes clusters have sufficient resources when demand increases.

How Stackbooster.io Up-scaling Works

Stackbooster.io uses a multi-faceted approach to determine when and how to scale up your Kubernetes clusters:

Predictive Scaling

Unlike standard Kubernetes autoscaling that reacts to resource saturation, our predictive scaling:

• Analyzes historical usage patterns to anticipate needs
• Identifies cyclic patterns (daily, weekly, monthly)
• Preemptively adds capacity before it's critically needed
• Reduces application lag caused by reactive scaling

Multi-Signal Analysis

Our platform considers multiple signals when making scaling decisions:

• CPU Utilization: Beyond simple thresholds, we analyze trends and rate of change
• Memory Usage: Both at node and pod levels, including projected memory growth
• Pod Scheduling Events: Failed scheduling due to resource constraints
• Application Metrics: Custom metrics from your applications when configured
• External Factors: Scheduled events, deployments, and known traffic patterns

Smart Instance Selection

When scaling up, our algorithm selects the optimal instance types:

• Analyzes the specific workload resource needs
• Considers price-performance ratio across instance families
• Evaluates availability of Spot instances for non-critical workloads
• Makes placement decisions based on workload affinity and anti-affinity

Configuring Up-scaling

Basic Configuration

To set up basic up-scaling parameters:

1. Navigate to your cluster in the Stackbooster.io dashboard
2. Select "Scaling Configuration" > "Up-scaling"
3. Configure the following settings:
   • Headroom Percentage: Buffer capacity to maintain (default: 15%)
   • Max Scale-Up Rate: Maximum nodes to add in a single scaling action
   • Predictive Scaling: Enable/disable predictive scaling
   • Response Sensitivity: How quickly to respond to increased resource demands

Advanced Settings

For more granular control, you can configure:

Workload-Specific Settings

Configure different scaling parameters for various workloads:

1. Navigate to "Workload Settings"
2. Select a workload or namespace
3. Configure custom scaling parameters:
   • Priority level for resource allocation
   • Minimum guaranteed resources
   • Maximum scale factor

Time-Based Rules

Create rules that modify scaling behavior based on time:

1. Go to "Scheduling Rules"
2. Create a new rule with:
   • Time window (e.g., business hours, weekends)
   • Custom headroom settings for the defined period
   • Special handling for known high-traffic events

Custom Metrics

Integrate application-specific metrics for scaling decisions:

1. Navigate to "Custom Metrics"
2. Configure your Prometheus or CloudWatch metrics
3. Set thresholds and scaling factors based on these metrics

Up-scaling Strategies

Stackbooster.io offers several up-scaling strategies to choose from based on your needs:

Balanced (Default)

• Maintains moderate headroom to handle normal traffic fluctuations
• Scales gradually to avoid over-provisioning
• Balances cost and performance considerations

Performance-Focused

• Maintains higher headroom for rapid response to traffic spikes
• Scales more aggressively when demand increases
• Prioritizes application performance over cost optimization

Cost-Focused

• Maintains minimal headroom to maximize resource efficiency
• Scales more conservatively, potentially accepting some scheduling delays
• Prioritizes cost savings over immediate scaling

Custom

• Define your own parameters for headroom, scaling speed, and instance selection
• Create different strategies for different environments or workloads
• Implement special handling for specific use cases

Best Practices

Determining Optimal Headroom

The ideal headroom percentage depends on your workload characteristics:

• Stable, predictable traffic: 10-15% headroom is typically sufficient
• Variable traffic with occasional spikes: 20-25% provides better responsiveness
• Highly unpredictable workloads: 30%+ may be necessary to prevent performance issues

Handling Planned Traffic Increases

For known traffic events (marketing campaigns, product launches, etc.):

1. Create an "Event Schedule" in Stackbooster.io
2. Define the expected traffic increase and duration
3. The platform will automatically adjust scaling parameters during the event

Monitoring Up-scaling Performance

To ensure your up-scaling is working effectively:

1. Monitor the "Scaling Performance" dashboard
2. Pay attention to metrics such as:
   • Average time to scale up when needed
   • Resource utilization after scaling events
   • Application performance during scaling transitions
3. Adjust your configuration based on observed patterns

Troubleshooting

Common Up-scaling Issues

Slow Response to Traffic Spikes

If your cluster isn't scaling up quickly enough:

• Increase the headroom percentage
• Check for AWS service quotas limiting instance launches
• Verify the instance types selected are readily available in your region
• Consider enabling predictive scaling if not already active

Excessive Resources After Scaling

If scaling up results in too much unused capacity:

• Reduce the headroom percentage
• Decrease the maximum scale-up rate
• Review instance selection to choose more appropriate sizes
• Consider more granular workload-specific scaling settings

Failed Node Launches

If nodes fail to launch during scaling:

• Check AWS service health in your region
• Verify IAM permissions for the node instance profile
• Review node launch templates for configuration errors
• Ensure you haven't reached AWS account limits

Advanced Topics

Combining with Horizontal Pod Autoscaling (HPA)

Stackbooster.io works in harmony with Kubernetes HPA:

• HPA adjusts pod counts based on metrics
• Our platform ensures nodes are available to accommodate those pods
• Configure both systems with compatible thresholds for optimal results

Custom Resource-Based Scaling

For specialized workloads with unique resource requirements:

1. Define custom resource metrics in the dashboard
2. Set thresholds for these metrics
3. Configure scaling actions based on these thresholds

Multi-Cluster Scaling Coordination

If you operate multiple clusters with interdependencies:

1. Use the "Cluster Groups" feature to define relationships
2. Configure coordinated scaling policies
3. Ensure resource allocation across the entire application landscape

By implementing these up-scaling strategies, your Kubernetes clusters will be able to handle increasing demand efficiently while maintaining optimal resource utilization and cost management.