This is the most expensive step that companies have to pay for when migrating their workload to cloud. Most companies have monolithic websites and bulk applications running on-premise. The architecture of these applications really make it very expensive to run the workload on cloud. There is no use in migrating the mammoth monolithic application to cloud. It won’t save money. Enterprises should re-architect their workload to run around a set of micro-services. These microservices must be very light weight that it has clear separation between computing and storage. It should not use local storage to store even log files and it should use cloud storage to store files uploaded by the users. 

The services should be so light weight that it can run on a container in a stateless mode while the state is maintained in specialized cloud services or databases. eg: session, cache can be moved to a different component in the cloud instead of keeping it sticky inside the container.

Not all layers or components of the application need to run on cloud. You can run your workload on the client browser. Today’s browsers are very powerful no matter whether they run on mobile or tablet or desktop. In a typical MVC application all the layers Model-View-Controller run on the server and the browser get all the HTML tags and file ready to render from the server. In the modern architecture you can decouple the M,V and C and then build views using client side framework like Angular, Vue or React and shift much of the client side workload to run on the browser. The controller can just be a set of API that the Angular can call and get/post the data and validate it and save it to the database. This forces many companies to learn JavaScript or TypeScript and this learning curve daunts some companies. This shift from monolithic MVC to component based rendering requires significant rearchitecting of your application.

The whole devops and testing processes need to be migrated to cloud. The non-functional components like security, logging, performance monitoring etc need to be moved to cloud as well. Cloud providers have specific set of services to for these non-functional requirements and the apps need to use them. The CI/CD pipeline needs to be rebuilt in cloud. If you have used standard set of tools instead of custom scripts then it may be easy to migrate them to cloud. The testing process should also be designed to be light weight that it does not require a dedicated server to run the test programs.

If the resource usage for some of your applications does not grow exponentially or if you can project the resource usage for future then you can go for committed use where you sign long term resource usage agreement to get the processing power, storage and networking at discounted price. Cloud computing companies provide you this committed use discount to buy only the processing power or storage or networking requirements or combination of these. Google cloud provides a very helpful tool that will recommend committed use that you can make use of to save money.