Our cost model allows user to distribute the their costs bassed of the CPU or memory usage of their user projects.
During data processing we add in additional rows to represent the worker unallocated cost. These rows are added under the project name Worker unallocated during the OCP daily summary table creation. Formula:
unallocated_worker_cost = capacity - effective usage
The effective usage is the max(request, usage), since users can have a higer usage than their request.
One of the features of our cost models is to distribute the cost associated with the Worker unallocated project to the user projects with this algorithm:
distributed_cost = (user_project_usage / usage_of_all_user_projects) * worker_unallocated_cost
For example, there are worker unallocated cost of 100 dollars and two user projects, A and B. Project A has a usage of 25 and project B has a usage of 75. Using the algorithm above to distribute the worker unallocated cost:
worker_unallocated_cost = 100
Project's A distributed_cost = (25 / sum(25 + 75)) * 100 = 25
Project's B distributed_cost = (75 / sum(25 + 75)) * 100 = 75
Project A gets an additional 25 dollars and Project B gets an additional 75 dollars due to distributing the worker unallocated cost.
How we deduplicate
The Worker unallocated project that we created earlier in the How We Calculate section still holds the original cost. Using the example above this would be our original 100 dollars. In order to deduplicate costs for the Worker unallocated project we need to subtract the unallocated costs:
distributed_cost = 0 - worker_uanllocated_cost
Working with our same example:
Worker unallocated project's distributed_cost = 0 - 100