Abstract

The increasing reliance on computing systems has greatly impacted the customers' expectations from such systems; for which the need for reliable and highly available services has become an essential requirement. This has led service providers to seek for new ways to supply robust services in order to sustain their advantage in today's highly competitive markets. A highly available system is defined as a system that is up and running 99.999% of the time. To gain such availability, a solution that has been widely employed is the usage of redundant components. However, solutions used in the past are proprietary and as a result applications had no portability from one platform to another. The Service Availability Forum (SAF), a consortium of several computing and telecommunication companies, promotes standardized solutions for building highly available systems in which Commercial-Off-The-Shelf (COTS) building blocks can be used. The SAF middleware has many services each of which has a specialized role so that the whole middleware can manage the redundant components within the system to fulfill the service availability. Like any other system, a SAF system may go through several upgrade and downgrade processes during its lifetime. Though due to the high availability requirement, a SAF system ought to be upgraded while ensuring minimum service interruption. Among the SAF middleware services, the Software Management Framework (SMF) is responsible for this live upgrade. In order for the SMF to perform an upgrade the road map of this migration, which is known as the Upgrade Campaign Specification, should be provided. However, due to the number of entities involved in an upgrade campaign and the complexity of the relationships among these entities, manual calculation of various steps of an upgrade campaign specification is time consuming and error prone. In this thesis, we have devised an approach for automatic generation of upgrade campaign specifications to upgrade redundant entities of SAF systems. We have categorized possible upgrade variations into three main scenarios which consist of manipulating current entities of the system, removing or adding new ones. For each scenario we have recognized different criteria that impact the service availability. For each criterion, according to the different upgrade methods introduced by SMF, we have devised solutions to minimize the service availability interruption during the course of an upgrade. Finally, we have created a prototype tool that supports the generation of upgrade campaign specification algorithms for each scenario. We have applied our approach to a case study to demonstrate its applicability.