Skip to main content

Features

A detailed look at what stepscale AI brings to your autoscaling infrastructure.

Auto-Tuning​

The core capability of stepscale AI. Instead of manually configuring scaling parameters and hoping they work, the AI continuously learns from your actual workload.

What gets tuned:

  • Scale-up threshold - the queue depth that triggers adding more tasks
  • Scale-down threshold - the queue depth that triggers removing tasks
  • Min/max task counts - adjusted per time-of-day and day-of-week patterns
  • Tasks-per-message ratio - calibrated to your actual message processing times

How it works:

  • Collects rolling metrics (queue depth, task count, processing rates) over 24-48 hour windows
  • Identifies patterns: daily cycles, weekly variations, burst characteristics
  • Calculates optimal parameter values using statistical analysis
  • Writes updated configuration that the reactive scaler picks up automatically

The AI analysis runs periodically (not on every scaling cycle), keeping compute costs minimal.

Anomaly Detection​

Not all traffic spikes are the same. Your daily 9am rush is predictable - a sudden spike at 3am is an incident.

Capabilities:

  • Classifies traffic patterns as normal or anomalous based on historical baselines
  • Applies different scaling strategies for anomalies (more aggressive scale-up, slower scale-down)
  • Integrates with alerting systems to notify your team when unusual patterns are detected
  • Learns from your response to past anomalies to improve future detection

Cost Optimization Insights​

The feature that pays for itself. stepscale AI provides actionable reports on your autoscaling efficiency.

What you get:

  • Over-provisioning detection - "Your min_tasks is 10 but you average 0.2 messages/sec between midnight and 6am"
  • Before/after comparisons - track how AI-tuned configurations perform vs your manual settings
  • Monthly savings estimates - concrete dollar amounts showing the impact of optimization
  • Right-sizing recommendations - when your current task size is too large or too small for your workload

Multi-Platform Support​

stepscale AI works across orchestration platforms, not just one.

Supported platforms:

  • AWS ECS - via Fast Autoscaler or native ECS autoscaling
  • Kubernetes - via HPA (Horizontal Pod Autoscaler) or KEDA

The AI tuning layer is platform-agnostic - it analyzes metrics and outputs configuration values. The platform-specific adapter translates those values into the right API calls.

Architecture​

Queue Providers           stepscale AI              Compute Providers
─────────────────        ─────────────────          ─────────────────
SQS                      Metrics Collection         ECS
Kafka            ───►    Pattern Analysis    ───►   Kubernetes
RabbitMQ                 Config Optimization
Redis                    Anomaly Detection
Kinesis                  Cost Reporting

The reactive scaler handles real-time decisions. stepscale AI handles the intelligence layer - analyzing, learning, and optimizing over time.