Advanced Scaling Strategies for High Availability

Cloud bursting

Used when an on-premises data center reaches saturation or capacity limits, either due to physical constraints or a lack of recent server provisioning. In this scenario, additional capacity is added in a public cloud infrastructure to handle the overflow of work

Oversubscription

Primary used for cost efficiency. It involves provisioning virtual resources that collectively exceed the available physical resources. This approach assumes that saturation will not occur, making it a common pattern for public cloud providers. Can be applied to CPU, Memory, and network throughput

Hypervisor Affinity

Prioritizes placing resources physically and virtually close together to achieve high network throughput and low latency. This is suitable for scenarios where most traffic occurs between nodes of a workflow, and resilience is not a top priority. It poses a risk of simultaneous failure if resources are clustered too closely (e.g., within the same rack during maintenance)

Hypervisor anti-affinity

Ensures that resources are separated physically, at least within separate racks or even different data centers or regions. This approach is suitable for critical instances that must remain independent of each other to maximize availability and resilience, minimizing the risk of simultaneous outages