Compute best practices

Compute refers to services that help in the execution of services. Some of the best practices related to compute are as follows:

• Choose the best location for your compute services, such as virtual machines. Choose a location where all Azure features and resources are available together in the same region. This will avoid egress traffic. 
• Choose the optimal size for your virtual machines. A bigger virtual machine costs more than a smaller one, and a bigger virtual machine might not be required at all.
• Resize virtual machines according to demand. Azure releases new virtual machine sizes frequently. If a new size becomes available that is better suited to your needs, then it should be used.
• Shut down compute services when they are not needed. This particularly applies to non-production environments.
• Deallocate virtual machines rather than shutting them down. This will release all resources and consumption will stop.
• Use development/testing labs for development and testing purposes. They provide policies and auto-shutdown and auto-start features.
• With virtual machine scale sets, provision few virtual machines and increase their count based on demand.
• Choose the correct size (whether small, medium, or large) for application gateways. They are backed up by virtual machines and can help reduce costs if sized optimally. Also, choose the basic tier application gateway if a web application firewall is not needed.
• Choose the correct tiers for virtual private network gateways (including a basic virtual private network, standard, high performance, and ultra performance).
• Minimize network traffic between Azure regions by collocating resources in the same region. 
• Use a load balancer with a public IP to access multiple virtual machines rather than assigning a public IP to each virtual machine.
• Monitor virtual machines and their performance and usage metrics. Based on those metrics, determine whether you want to upscale or scale out the virtual machine. Consultation of the metrics could also result in downsizing the virtual machines.