Towards Continuous Delivery by Reducing the Feature Freeze Period: A Case Study

Today, many software companies continuously deliver and deploy new features to their customers. However, many software systems are still released traditionally with long feature freeze periods and time-based releases due to historical reasons. Currently, only a few empirical inquiries of transformations towards continuous delivery exist. In this paper, we aim to understand how feature freeze was practiced and the feature freeze period reduced in an R&D program at Ericsson. The case organization has struggled with the feature freeze approach and is now moving towards the continuous delivery paradigm. We investigated the intended and actual effects of the feature freeze practice, how the feature freeze period was reduced and what effects the reduction had. We interviewed 11 employees, covering all the development teams at the largest site of the distributed organization. In addition, we analyzed data from software repositories to get quantitative triangulation of the qualitative results. Historically, the organization was not able to comply with the intended feature freeze practice, due to pressure for new feature development and long feature freeze periods leaving little time to perform actual development. By implementing test automation, the organization was able to reduce the feature freeze period by 56%, after which the amount of changes during the freeze decreased by 63% and the amount of changes close to the release date by 59%. We conclude that reducing the feature freeze period is possible using test automation, and reducing the freeze time can increase conformance to the intended feature freeze practice. To further reduce feature freeze, attention must be paid to deployment automation and collaboration between development and operations, in addition to test automation.