How do you chargeback or any recommendation of an application running in Cloud native K8s or ARO cluster with ideal charges baked into it proportionately?
Sort By:
Oldest
VP of IT in IT Services8 months ago
Without knowing much about how the application is used and if the cluster is shared, I'll assume it's dedicated. Average consumption (CPU and memory) will be the start. The reserves to cater for the peak will follow to be added along with the support and maintenance of the infrastructure/DBs too. So add them into the CPU prices but this will become hard to explain the deviation from the underlying principle provide. Another option will be to call it uplifts (for internal) to cater for the above. Chief Technology Officer in Software8 months ago
There is quite a lot to this question - much too much to write it all here but this article does a pretty good job of explaining the approach with K8s. https://blog.kubecost.com/blog/kubernetes-chargeback/CTO for Digital & IT in Healthcare and Biotech8 months ago
To add to what some of the others have already answered, our infrastructure department has approached this by essentially creating separate K8S clusters, one for each of its customers, which are the business IT departments in charge of applications. So that way, business IT knows that it is paying the infrastructure group for exactly what it is using, and is free to allocate costs among its applications however it wants (and is of course provided some metrics to ensure this can be done in a reasonably fair manner) but business A doesn't end up paying for business B's infrastructure.Director of IT in Manufacturing8 months ago
Hi Jeremy, Actually it was an one of the option initially. But some customers are unwilling to pay the price of a one complete dedicated cluster cost to host one single application where application usage is not that significantly high that can consume entire cluster's resources.
Thanks for the reply.
IT & Strategy Advisor to CxOs || Digital & Enterprise Architecture Consultant in Consumer Goods8 months ago
Adding to what others have recommended. The chargeback mechanism depends on the models for setting up clusters and resource sharing between teams/businesses. For models such as dedicated clusters for each team/business or dedicated node group in a shared cluster for a team/business, the most important step is to set up Kubernetes cost monitoring mechanisms to get insight into cost structure. Also, ensuring that cloud vendors provide a detailed billing that will help to calculate cost of running K8 resources such as clusters and nodes.Next, organisations also need to consider chargeback for shared or idle K8s costs, which is unavoidable. One is always challenged to arrive at an appropriate model to allocate these costs. Here are some models used by organisations -
1) Kubernetes costs are shared equally with all the tenants. For example if a cluster has 2 namespaces, each allocated to one team/business, the shared or idle costs are charged to both the teams equally.
2) Based on resource allocation. For example, namespace # 1 is allocated 50%, namespace # 2 is allocated 25% and namespace # 3 is allocated 25%, the tenant using 50% contributes to the largest share of cost (50%), irrespective of it being used or idle, and the other two will 25% each of shared or idle costs.
3) Some organisations divide shared/idle costs of K8s between business and IT, depending upon the IT's ability to absorb the costs.
When the organisations allocate namespaces in a shared cluster, it becomes more complex as it will require a very granular level of resource utilization metrics.
Further, many organisations offer a catalog-driven approach for chargeback. However, it depends upon organisation culture, maturity and data-driven approach to build the catalog and implement it.
Hope it helps.