Below, we will use the terminology used by Amazon Web Services to detail the elements that make up an autoscaling system, without mentioning the specific names of the services provided by the company. Alternative names used in other platforms (Google Cloud, Microsoft Azure and others) are also mentioned.

Amazon Web Services (AWS)

Amazon Web Services launched the Amazon Elastic Compute Cloud (EC2) service in August 2006, which allowed developers to programmatically create and terminate instances (machines).[7]​[8]​ At the time of launch, AWS did not offer auto-scaling, but the ability to create and terminate instances this way gave developers the flexibility to design their own scaling strategy.

Third-party software for autoscaling appeared around April 2008. These included tools from Scalr[9]​ and RightScale. Animoto used RightScale, which was able to handle Facebook traffic by adopting this technique.[10]​[11]​.

On May 18, 2009, Amazon launched its own auto-scaling feature alongside Elastic Load Balancing as part of Elastic Compute Cloud, which became an integral component of the service.[12]​ Configuration can be done through a web browser or the command-line utility (aws-cli).[13]​ As of May 2016, auto-scaling is also offered for the AWS ECS service.

Microsoft Azure

On June 27, 2013, Microsoft announced the addition of auto-scaling support to its Microsoft Azure platform.[14]​[15][16]​ Official documentation is available on the Microsoft Developer Network.[17][18]​.

Oracle Cloud Platform

Oracle Cloud Platform allows server instances to automatically scale a cluster in or out by defining rules.[19]​ These rules are based on CPU and/or memory utilization, and determine when to add or remove nodes.

Google Cloud Platform

On November 17, 2014, Google Compute Engine announced a public release of its auto-scaling feature for use in Google Cloud Platform applications.[20][21][22][23]​.

Facebook

In an August 2014 post, a Facebook engineer revealed that the company had begun using autoscaling within its own infrastructure to reduce its energy costs. The company reported a 27% decrease in energy use for off-peak hours (around midnight) and a 10-15% decrease during the typical 24-hour cycle.[2][24]​.

Horizontal Pod Autoscaling in Kubernetes

Kubernetes Pod Horizontal Autoscaling automatically scales the number of pods in a replication controller, deployment, or replica set based on observed CPU utilization (or, with beta support, some other metric provided by the application).[25]​.

Netflix

An example of the great benefits of using auto-scaling in high traffic consumption companies is Netflix, the video-on-demand provider, who documented how using this technique helped meet their highly variable consumption needs.[6]​.

Contenido

El autoescalado por defecto usa un enfoque reactivo para tratar con el escalado del tráfico: el escalado sólo ocurre en respuesta a cambios en tiempo real en las métricas. En algunos casos, particularmente cuando los cambios ocurren demasiado rápido, el enfoque reactivo puede resultar insuficiente. A continuación se describen enfoques alternativos.

Scheduled autoscaling approach

In this approach, changes to the minimum size, maximum size, or desired capacity are scheduled at specific times of the day. Scheduled scaling is useful, for example, if the increase or decrease in load at certain times is known in advance, but the changes are so sudden that reactive autoscaling fails to respond in time.

Predictive autoscaling approach

This approach uses predictive analytics. The idea is to combine recent usage trends with historical usage data, as well as other types of data, to predict usage in the future, and automatically scale based on these predictions.

For example, for specific parts of its infrastructure and workloads, Netflix found that Scryer, its predictive analytics engine, gave better results than Amazon's reactive auto-scaling approach. In particular, it was better for:[26][24]​.

• - Identify large demand peaks and prepare capacity with a minimum of advance notice.

• - Deal with large-scale outages, failure in full availability zones and regions.

• - Deal with variable traffic patterns, providing more flexibility in the rate of horizontal or vertical scaling based on the typical level and rate of change in demand at various times of the day.

• - Cloud computing.

• - Load balancing.

• - Amazon Web Services.

• - Auto Scaling with Docker.

Below, we will use the terminology used by Amazon Web Services to detail the elements that make up an autoscaling system, without mentioning the specific names of the services provided by the company. Alternative names used in other platforms (Google Cloud, Microsoft Azure and others) are also mentioned.

Amazon Web Services (AWS)

Amazon Web Services launched the Amazon Elastic Compute Cloud (EC2) service in August 2006, which allowed developers to programmatically create and terminate instances (machines).[7]​[8]​ At the time of launch, AWS did not offer auto-scaling, but the ability to create and terminate instances this way gave developers the flexibility to design their own scaling strategy.

Third-party software for autoscaling appeared around April 2008. These included tools from Scalr[9]​ and RightScale. Animoto used RightScale, which was able to handle Facebook traffic by adopting this technique.[10]​[11]​.

On May 18, 2009, Amazon launched its own auto-scaling feature alongside Elastic Load Balancing as part of Elastic Compute Cloud, which became an integral component of the service.[12]​ Configuration can be done through a web browser or the command-line utility (aws-cli).[13]​ As of May 2016, auto-scaling is also offered for the AWS ECS service.

Microsoft Azure

On June 27, 2013, Microsoft announced the addition of auto-scaling support to its Microsoft Azure platform.[14]​[15][16]​ Official documentation is available on the Microsoft Developer Network.[17][18]​.

Oracle Cloud Platform

Oracle Cloud Platform allows server instances to automatically scale a cluster in or out by defining rules.[19]​ These rules are based on CPU and/or memory utilization, and determine when to add or remove nodes.

Google Cloud Platform

On November 17, 2014, Google Compute Engine announced a public release of its auto-scaling feature for use in Google Cloud Platform applications.[20][21][22][23]​.

Facebook

In an August 2014 post, a Facebook engineer revealed that the company had begun using autoscaling within its own infrastructure to reduce its energy costs. The company reported a 27% decrease in energy use for off-peak hours (around midnight) and a 10-15% decrease during the typical 24-hour cycle.[2][24]​.

Horizontal Pod Autoscaling in Kubernetes

Kubernetes Pod Horizontal Autoscaling automatically scales the number of pods in a replication controller, deployment, or replica set based on observed CPU utilization (or, with beta support, some other metric provided by the application).[25]​.

Netflix

An example of the great benefits of using auto-scaling in high traffic consumption companies is Netflix, the video-on-demand provider, who documented how using this technique helped meet their highly variable consumption needs.[6]​.

Contenido

El autoescalado por defecto usa un enfoque reactivo para tratar con el escalado del tráfico: el escalado sólo ocurre en respuesta a cambios en tiempo real en las métricas. En algunos casos, particularmente cuando los cambios ocurren demasiado rápido, el enfoque reactivo puede resultar insuficiente. A continuación se describen enfoques alternativos.

Scheduled autoscaling approach

In this approach, changes to the minimum size, maximum size, or desired capacity are scheduled at specific times of the day. Scheduled scaling is useful, for example, if the increase or decrease in load at certain times is known in advance, but the changes are so sudden that reactive autoscaling fails to respond in time.

Predictive autoscaling approach

This approach uses predictive analytics. The idea is to combine recent usage trends with historical usage data, as well as other types of data, to predict usage in the future, and automatically scale based on these predictions.

For example, for specific parts of its infrastructure and workloads, Netflix found that Scryer, its predictive analytics engine, gave better results than Amazon's reactive auto-scaling approach. In particular, it was better for:[26][24]​.

• - Identify large demand peaks and prepare capacity with a minimum of advance notice.

• - Deal with large-scale outages, failure in full availability zones and regions.

• - Deal with variable traffic patterns, providing more flexibility in the rate of horizontal or vertical scaling based on the typical level and rate of change in demand at various times of the day.

• - Cloud computing.

• - Load balancing.

• - Amazon Web Services.

• - Auto Scaling with Docker.