Spotinst Elastigroup will soon be celebrating its 4th birthday, and this is definitely a good reason for celebration.
Since the product was initially launched we have added many different capabilities and integrations in order to supply our customers with a platform which is more robust, and that can handle various types of workloads when running operations in the cloud.
Today we are proud to announce Predictive Auto-Scaling for Elastigroup, powered by Machine Learning, data collected across billions of data points and trends we analyzed in the past years, this feature will empower the ongoing automation of our customers’ cloud infrastructure.
- Forecast – The minimum amount of instances an Elastigroup should have in order to accommodate the load required by the predicted metric.
- Predicted Metric – The metric which Elastigroup actively predicts, in order to determine future load and scaling requirements for the cluster.
The elasticity of the cloud has enabled the fundamental ability to scale instances up or down based on a given load of an application.
The daily motivation here at Spotinst is making the lives of our customers much easier, therefore, we have invested a lot of time and effort in perfecting our auto-scaling mechanism.
Elastigroup Predictive Auto-Scaling is simplifying the process of defining scaling policies, identifying peak times in the platform, and automatically scaling the right capacity in advance.
System Process Flow
In order to manage the various technological challenges in the prediction mechanism, we developed Machine Learning algorithms which collect and process the cluster’s metrics over time (CPU and In\Out Network Traffic), in order to create a baseline which will eventually provide a predicted metric of the cluster’s future load.
Once the predicted metric value of the elastigroup is determined, a calculation begins to determine the minimum amount of instances that are required to handle the predicted load of the cluster.
The prediction mechanism of the auto-scaling was developed based on the prediction algorithm of Spotinst’s spot-interruption detection.
In addition to that, the algorithm knows to differentiate between normal cycles and anomalies, which can incorrectly scale your cluster up\down.
The cycle learning period of Predictive Auto-Scaling is based on the timeframe in question, meaning:
- For an hourly forecast, the algorithm will need a 1 day learning period in order to predict the following day’s load at the same hour.
- For a daily forecast, the algorithm will need a 1 week learning period in order to predict the following week’s load at the same day.
Predictive Auto-Scaling Configuration
Under the Elastigroup scaling configuration, when selecting “Target Based Scaling” you will have a checkbox that activates the predictive auto-scaling feature.
Currently, the supported metric for Predictive Auto-Scaling is CPU Utilization of the Elastigroup cluster.
Once you have selected the checkbox, you can choose from a drop-down list which predictive auto-scaling functions you wish to apply to your cluster.
Spotinst offers 2 functions for Predictive Auto-Scaling:
- Predict Only
- Predict and Scale
Selecting either one of these functions will start the metric prediction process, which is empowered by machine learning models that determine the expected target metric value , based on the data collected from previous behavior.
Predict Only Mode
When selecting this function, Elastigroup will present its predicted data, but will scale instances normally, according to the regular “Target Based Scaling” behavior.
This function allows you to analyze the predicted metric values, without automatically scaling the cluster according to these values.
These values will be presented as predicted metrics in a graph view, as part of the “Overview” tab of your Elastigroup.
You may find this function useful when first approaching the predictive auto-scaling feature.
Predict & Scale Mode
When selecting this function, Elastigroup will not only present its predicted data, but will also scale the cluster in order to meet the Forecast minimum amount of instances that will accommodate the predicted load.
Elastigroup will automatically scale up instances according to the calculated Forecast value, relieving you from the hustle of attempting to predict your cluster needs in advance.
Predictive Auto-Scaling is now available for all Elastigroups in all regions.
Save time and effort and let Spotinst predict future loads and automatically scale your cluster accordingly.
Ready to rock n’ roll? Make the most out of your Elastigroup cluster!