Predictive Rebalancing

What is Predictive Rebalancing?

Predictive Rebalancing is based on a machine learning model used for identifying and predicting spot capacity usage and interruptions in the public cloud. Its purpose is to predict which markets will experience interruptions and replace the instances within these markets before the interruption. This prevents under-capacity and ensures availability of resources.

The rebalancing model is application–driven, combining advanced predictive algorithms with an understanding of infrastructure and real-time analysis of workload requirements.

The result is a process that takes the unique requirements of each workload into consideration to automatically provision and allocate cloud infrastructure in the most efficient and effective way possible.

Elastigroup provides a centralized view of capacity, optimization and availability for meeting your specifications to allow for complete control over your Elastigroup settings:


Instance Availability

This option enables you to define availability preferences and gain visibility into EC2 spot instance availability and lifespan. Elastigroup can be configured to honor the following criteria for instance availability:

  • Minimum Instance Lifetime: The desired amount of time you wish your workloads to run without any interruption to their underlying instances. The shorter the lifetime you choose, the more accurate the market scoring will be.
  • Preferred Availability Zones: Choose one or more of the availability zones that you defined in the Compute tab.
  • Preferred Spot Types: Choose one or more of the instance types you defined in the Compute tab.
  • Draining Timeout: In addition, you can define the draining period your application requires so the automation will start replacing the instances with enough time before the interruption is predicted to occur, allowing for complete and graceful draining.