QoS Accoutability Management in SOA Project

Service-Oriented Architecture (SOA) provides a powerful paradigm to dynamically compose service processes using individual atomic services. It benefits business in terms of rapid adaption to the changing conditions and requirement, as well as the maximal reuse of IT assets.

Accountability in SOA aims to provides a clear sense of which services are responsible for the undesired outcome of process at run-time. Without accountability, the service process execution appears as a black box to users.

It is challenging to acheive accountability in SOA because these service processes can become vastly complex, involving many service partners, thereby giving rise to difficulties in terms of run-time problem detection and diagnosis. Moreover, in SOA outputs from individual services have dependencies: a low output quality from a service may cause the output quality degradation of all its successors in a service process. Even worse, successive quality degradation may be accumulative or multiplicative. Therefore, the root causes of a problem and their consequences must be clearly distinguished to make SOA environment transparent, traceable, and accountable.

We present the design of an accountability framework as part of an integrated SOA deployment and management solution to be in charge of detect, diagnose, and defuse the root cause of a service process deficiency (such as functional errors or SLA violations).

The accountability framework provides the mechanisms for:

• Run-time process monitoring performed by agents, which are selected based on effective agent deployment strategies to minimize the monitoring infrastructure cost
• Bayesian network reasoning to identify the likely causes of a problem and to selectively inspect those services
• An evidence channel selection algorithm to find an optimized information collection structure
• Continual process reconfiguration and optimization
• A trust and reputation management network for prevention of future problems

More information can be found in the Publications page.