Abstract

Oussama Jebbar

High availability is an important non-functional requirement for carrier grade services. The applications/systems providing and protecting such services undergo frequent upgrades which makes meeting this requirement very challenging. A system upgrade is the migration process from the system’s current configuration to a new one. This migration may include configuration changes, installation and removal of software, etc.. The Service Availability Forum (SAF) published a set of standards that describe a high availability enabling middleware for Commercial-off-the-shelf (COTS) components based systems. In such a middleware, the Software Management Framework (SMF) is the service responsible for orchestrating the upgrades. These upgrades are performed according to a road map called an upgrade campaign specification. The Availability Management Framework (AMF) is another service defined in the SAF standards and which is responsible of managing the availability of the services and the service providers. To take a SAF compliant system from one configuration to another, one has first to come up with an upgrade campaign specification for that purpose. Moreover, there are multiple upgrade campaign specifications that can take the system from the same source configuration to the same target configuration, but they differ in the duration they take and the service outage they may induce. Designing an upgrade campaign specification for a SAF compliant system is not a straight forward process. Indeed, this is an error prone task that becomes more challenging when the system and the set of changes to perform get larger. Besides, selecting which upgrade campaign specification to apply among all the valid ones is either expensive (running the same upgrade campaign specification on a replica of the real system), or tedious (evaluating, comparing and selecting upgrade campaign specifications manually).
In this thesis we propose automation as a solution to ease and minimize human intervention in the design and evaluation of upgrade campaign specifications. We devise a model driven approach to automatically generate upgrade campaign specifications. Our approach consists of several activities in order to ensure the SAF compliance of the generated upgrade campaign specification on one hand. On the other hand, we use the dependencies between system components to apply a set of rules that can improve the quality of the upgrade campaign specification by avoiding some of the unnecessary service outage. These rules include rules to order changes to be performed as well as a set of heuristics that make use of the dependencies. We also address the upgrade campaign evaluation related issues. We extend an existing discrete event systems based simulation approach for upgrade campaign evaluation. We expose the limitations of a random simulation as its results are unreliable for comparison. To overcome these limitations we define best case and worst case scenarios that we use to guide upgrade campaign simulations to see how the upgrade campaigns perform in edge cases. We also devise a method for upgrade campaign specification selection/elimination based on applicability checks according to two criteria: the maintenance window, and the acceptable outage during this window. Finally, we implemented prototypes for upgrade campaign specification generation and evaluation.