Higher availability with Scheduled preventive replacement of Spot Instances

In the past 4 years, we’ve been tirelessly working to develop creative ways to leverage cloud computing excess capacity for production and mission-critical applications using Spotinst Elastigroup.

Prior to Spotinst Elastrigroup, users were very hesitant to leverage Amazon EC2 Spot Instances for several types of applications, let alone that the market’s awareness was relatively low. The main reason for this hesitation relates to the nature of Spot Instances which is manifested by an unknown lifespan, and that eventually they will be terminated with a 2-minute notification, AKA “Interruption”.

Since we have begun the development of Elastigroup, we have been working very hard on collecting the interruption data, analyzing it and cross-referencing it with real-time data sources in order to provide reliable conclusions. Analyzing the interruptions has provided us with the visibility to identify valuable patterns and regularity, and those allowed us to be rapid and agile in replacing spot instances prior to their interruption schedule by AWS. As time went by we have excelled in predicting interruptions and reached a prediction of up to 15 minutes prior to their occurrence, which allows a more spacious infrastructure with high availability.  

Today, We are excited to announce, that we have recently strengthened Spotinst Elastigroup’s preventive replacement mechanism with a new capability called “Scheduled preventive replacement”.  Whereas in parallel to the data-driven prediction method, this new capability will allow Elastigroup to proactively replace EC2 instances in direct adequacy with the average lifespan of that particular instance. This means that the prediction algorithm will analyze the average runtime of the EC2 instance type in the relevant AZ, and replace it with a newer instance, and therefore reducing the interruption risk dramatically.

Scheduled preventive replacement is embedded into Elastigroup’s prediction algorithm, so no configuration is required from the user. This capability is available and already assuring highly available workloads for our customers.