Chandra Science Operational Data System Migration to Linux: Herding Cats through a Funnel

Migration to a new operational system requires technical and non-technical planning to address all of the functional associations affiliated with an established operations environment. The transition to (or addition of) a new platform often includes project planning that has organizational and operational elements. The migration likely tasks individuals both directly and indirectly involved in the project, so identification and coordination of key personnel is essential. The new system must be accurate and robust, and the transition plan typically must ensure that interruptions to services are minimized. Despite detailed integration and testing efforts, back-up plans that include procedures to follow if there are issues during or after installation need to be in place as part of the transition task.

In this paper, we present some of the important steps involved in the migration of an operational data system. The management steps include setting objectives and defining scope, identifying stakeholders and establishing communication, assessing the environment and estimating workload, building a schedule, and coordinating with all involved to see it through.

We discuss, specifically, the recent migration of the Chandra data system and data center operations from Solaris 32 to Linux 64. The code base is approximately 2 million source lines of code, and supports proposal planning, science mission planning, data processing, and the Chandra data archive. The overall project took approximately 18 months to plan and implement with the resources we had available. Data center operations continued uninterrupted with the exception of a small downtime during the changeover. We highlight our planning and implementation, the experience we gained during the project, and the lessons that we have learned.