Service level agreement (SLA) is an essential agreement formed between a consumer and a provider in business activities. The SLA defines the business terms, objectives, obligations and commitment of both parties to a business activity, and in cloud computing it also defines a consumer's request for both fixed and variable resources, due to the elastic and dynamic nature of the cloud-computing environment. Providers need to thoroughly analyze such variability when forming SLAs to ensure they commit to the agreements with consumers and at the same time make the best use of available resources and obtain maximum returns. They can achieve this by entering into viable SLAs with consumers. A consumer's profile becomes a key element in determining the consumer's reliability, as a consumer who has previous service violation history is more likely to violate future service agreements; hence, a provider can avoid forming SLAs with such consumers. In this paper, we propose a novel optimal SLA formation architecture from the provider's perspective, enabling the provider to consider a consumer's reliability in committing to the SLA. We classify existing consumers into three categories based on their reliability or trustworthiness value and use that knowledge to ascertain whether to accept a consumer request for resource allocation, and then to determine the extent of the allocation. Our proposed architecture helps the service provider to monitor the behavior of service consumers in the post-interaction time phase and to use that information to form viable SLAs in the pre-interaction time phase to minimize service violations and penalties. ; No Full Text