Service-oriented architectures (SOA) provide enterprises the flexibility and cost-savings of dynamically composing business workflows from reusable services available anywhere remotely, potentially from third-party vendors. However, despite these new advantages, two aspects of SOA still remain a barrier to achieving end-to-end quality of service (QoS) guarantees for workflows: (1) in such a distributed setting, workflow behaviors are generally opaque such that it becomes difficult to identify the root causes of faults when they occur; and (2) current SOA frameworks do not provide the policies and mechanisms for deadline-driven workflow planning. Therefore, both enterprises and service vendors will need new tools to ensure the end-to-end QoS of workflows. To address opacity, enterprises will demand service accountability and transparency from service vendors. Likewise, service vendors will require the means to guarantee timely and predictable performance of their services.

To address these new requirements, this dissertation discusses how the Llama SOA middleware framework achieves accountable and predictable SOA both at the service-level, by enhancing existing open-source SOA middleware with monitoring, reconfiguration, and service execution reservation capabilities; and at the end-to-end workflow-level, by creating a distributed component infrastructure for problem detection, diagnosis, and deadline-based workflow composition.

Our experimental results indicate that using Llama's accountability infrastructure contributes only a modest amount of system overhead, and that the diagnosis process for pinpointing the root cause of process problems is swift and sufficiently accurate. Moreover, our evaluation of Llama's predictability framework demonstrates that service-level reservation management data structures are efficient with logarithmic time complexity over an indefinite time horizon, that a reserved service successfully executes on-time upon the arrival of its request, and that workflow reservations themselves can be made quickly and predictably by using a novel approach for achieving fast concurrent service reservations.