Resource Scaling and System expansion

In the process of cloud architecture design, it is crucial to consider the scalability and elasticity of the architecture to achieve high-performance cloud infrastructure. Different application deployment methods require different scaling solutions. The following are common approaches:

• Cloud Service Auto Scaling: Elastic scaling on Alibaba Cloud relies on Alibaba Cloud Auto Scaling product (ESS), also known as Auto Scaling. ESS is a service that automatically adjusts compute capabilities (i.e., the number of instances) based on business needs and policies. It currently supports elastic scaling of ECS and ECI instances. ESS has a wide range of application scenarios and is not only suitable for applications with fluctuating business volumes but also for applications with stable business volumes.

• Container Auto Scaling: With the growing adoption of containers as the core computation technology in cloud computing, more and more application systems have completed or are undergoing containerization and are running in various container environments. The corresponding container service on Alibaba Cloud is the Container Service for Kubernetes (ACK). Elastic scaling, in the above case, corresponds to typical scenarios like online business elasticity, large-scale computation training, deep learning GPU or shared GPU's training and inference, regular and periodical workload changes, etc. Elastic scaling comprises two dimensions:

• Scheduler layer elasticity: mainly responsible for modifying changes in the scheduling capacity of the workload. For example, the HPA is a typical component of the scheduler layer elasticity. By utilizing HPA, the number of replicas for the application can be adjusted. The number of replicas that were adjusted will change the scheduling capacity currently occupied by the load, thus achieving scaling at the scheduler layer.

• Resource layer elasticity: mainly used when the capacity planning of the cluster cannot meet the scheduling capacity of the cluster. In this case, the scheduling capacity will be replenished by elastically scaling out resources like ECS or ECI.

  Both flexibility components and capabilities can be used separately or in combination, and they are decoupled from each other via the scheduling layer's capacity status.

• Serverless: for users, both cloud services and container scaling require a certain level of involvement in managing and maintaining infrastructure-level resources. However, with Serverless technology, users are relieved of the need to be concerned about infrastructure-level scaling and capacity planning. Serverless-related cloud products offer three key features: elastic scalability (scaling in milliseconds), cost optimization (paying only for actual usage), and high development efficiency (no need to worry about underlying cloud resources). Up to now, Alibaba Cloud has developed more than 20 core cloud products in the Serverless architecture domain, providing comprehensive services ranging from Serverless computing to Serverless application development capabilities.