Q15Cloud Computing

It ensures that the right amount of resources is available to an application at the right time. Over-provisioning leads to wasted resources and higher costs, while under-provisioning leads to performance degradation and SLA violations. Cloud computing relies on dynamic provisioning, where resources are scaled up or down automatically based on workload demands, ensuring elasticity and optimal utilization.

Types of Provisioning

• Static (Advance) Provisioning: Resources are allocated in advance based on predicted demand, and the customer is billed a flat rate regardless of actual usage. This suits applications with predictable, steady workloads.
• Dynamic Provisioning: Resources are allocated and de-allocated automatically as demand fluctuates, and the customer is billed only for what is consumed. This suits workloads with variable or unpredictable traffic, such as e-commerce sites during sales events.
• User Self-Provisioning: The customer directly requests resources from the provider's portal or API and receives them almost instantly, without any provider-side manual intervention.

Effective provisioning relies on auto-scaling policies that monitor metrics like CPU utilization or request queue length and trigger the addition or removal of virtual machine instances, balancing performance guarantees against infrastructure cost. Providers such as AWS Auto Scaling and Azure Autoscale implement this by combining a monitoring service with predefined scaling policies, allowing the infrastructure to react to demand within minutes rather than requiring manual intervention.

Provisioning Trade-offs

The central engineering challenge in resource provisioning is balancing two opposing costs. Over-provisioning wastes money on idle capacity that never gets used, directly eroding the cost benefits that motivated the move to the cloud in the first place. Under-provisioning risks slow response times, dropped requests, and breached SLA commitments during traffic spikes, which can cause direct revenue loss for customer-facing applications such as e-commerce or streaming platforms. Predictive provisioning, which uses historical traffic patterns and machine learning forecasts to scale resources ahead of an anticipated spike (for example, before a scheduled sale event), is increasingly used alongside reactive auto-scaling to reduce the lag between a demand spike and the infrastructure's response to it.