Constraint-Based Runtime Prediction of SLA Violations in Service Orchestrations

Abstract

Service-Oriented Computing is an approach to creating applications where different loosely-coupled software services are composed in order to accomplish a goal that is more complex than what the constituent services can do. The composition is performed in a platform-independent manner (component services are executed in heterogeneous, usually non-local environments and accessed through a standardized interface) and is usually a long-running process which spans across organizations and administrative boundaries. In turn, these combinations are usually exposed as services themselves.Service combinations are usually divided into orchestrations and choreographies. In the former case there is a single agent which controls the individual services and routes the data between them. In the latter, data movements and control are not centralized. In this talk we will focus on service orchestrations.A critical point for the usability of service compositions is the Quality-ofService (QoS) they offer. Execution time, availability, or monetary cost are some usual metrics. The acceptable values for QoS attributes in a business relation are usually defined in Service Level Agreements (SLAs), along with the penalties in case they are violated.We present and evaluate a method whereby, using techniques from constraint logic programming, we derive, at a given point of execution of a service composition, a set of constraints that predict SLA conformance and violation scenarios over a certain time horizon. This is done on the basis of the structure of the composition and known or empirically measured properties of the component services. SLA failure and conformance constraints are expressed symbolically and may be used by other components for, e.g., development of data-mining models, optimized service matching, or triggering preventive adaptation or healing. Additional precision can be obtained within the same analysis framework by inspecting the state of the composition at the point of prediction.

A causal model to predict the effect of business process evolution on quality of service

“…If their approach is more accurate in terms of QoS prediction than ours, we are more time efficient since we reduce the number of elements of the system to recheck while they recheck the entire system. Ivanovic et al determine both at design time and runtime if Service Level Agreements (SLA) of an orchestration can be violated [10]. As us, they focus on QoS metrics that are measurable and can be aggregated.…”

A Constraint-Based Approach to Quality Assurance in Service Choreographies

“…To stay close to the executable specifications, we follow the same approach as in our previous work on run-time prediction for orchestrations [14]. We use the notion of a continuation which describes the current state of the participant and the remainder of the computation until its end [18].…”

“…We represent continuations using an abstract language for the participant processes ( Figure 3). It is based on a prototypical process language implementation that provides the continuation information explicitly at each execution step [14]. The participant state is kept in variables whose types are described in Section 3.4.…”

“…The constraint QoS model for a participant is derived automatically from the continuation and the previously accumulated QoS, using the structural approach [14], where QoS values for complex constructs are constructed from their structural components. A separate constraint QoS model is derived for each QoS metric of interest.…”

“…The proposed approach can be applied both at design time and at run time to support QoS negotiation, prediction, and QoS-driven adaptation. This work is an extension of [14] (on run-time prediction of SLA violations) for the case of service orchestrations with interconnected constraint models of stateful, interacting choreography participants, combining the derivation of QoS constraints with static analysis techniques.…”