Optimization Approach and Algorithm
Compute Software uses Machine Learning and a Wall Street-like computational approach that balances economic payoffs while accounting for the optionality of possible future decisions. Our algorithms make predictions and precise tradeoffs across provider, performance, and engineering costs.
It is through this approach that our solution can directly tie infrastructure spend to business outcomes, customize optimizations based on business objectives and workload characteristics, prioritize the most valuable actions for cloud optimization, and build trust with Engineering and IT organizations.
- Provider costs are costs attributable to cloud providers, such as Amazon Web Services, Microsoft Azure, and Google Cloud.
- Performance costs are costs attributable to non-perfect performance of your workloads and cloud resources. These costs are used to balance against provider costs for receiving optimal recommendations. The performance costs are based on the Business Criticality setting you choose and resource metrics (CPU, GPU, memory, storage, network, etc).
- Engineering costs are costs attributable to performing a task, like validating the task or resolving the task.
Recommendations are based on projections of provider, performance, and engineering costs. The recommendations are highly customizable.
As an example for compute rightsizing: by default, the performance cost for a potential rightsizing machine candidate is based on the resource metrics gathered in 10-minute intervals from the previous 14 days. When the projected metrics stay safely below 84% for each time interval, there is no performance cost. But as the metric values grow from there, the performance cost grows as well. (There are three predefined “Business Criticality” settings to easily customize the size of this performance cost growth.) The 10-minute performance costs are averaged and annualized, so that they can be balanced against the provider costs and the engineering costs of validating and resolving the rightsizing task.