Dragan Ivanovic

Comparing and Combining Predictive Business Process Monitoring Techniques

Predictive business process monitoring aims at forecasting potential problems during process execution before they occur so that these problems can be handled proactively. Several predictive monitoring techniques have been proposed in the past. However, so far those prediction techniques have been assessed only independently from each other, making it hard to reliably compare their applicability and accuracy. We empirically analyze and compare three main classes of predictive monitoring techniques, which are based on machine learning, constraint satisfaction, and Quality-of-Service (QoS) aggregation. Based on empirical evidence from an industrial case study in the area of transport and logistics, we assess those techniques with respect to five accuracy indicators. We further determine the dependency of accuracy on the point in time during process execution when a prediction is made in order to determine lead-times for accurate predictions. Our evidence suggests that, given a lead-time of half of the process duration, all predictive monitoring techniques consistently provide an accuracy of at least 70%. Yet, it also becomes evident that the techniques differ in terms of how accurately they may predict violations and nonviolations. To improve the prediction process, we thus exploit the characteristics of the individual techniques and propose their combination. Based on our case study data, evidence indicates that certain combinations of techniques may outperform individual techniques with respect to specific accuracy indicators. Combining constraint satisfaction with QoS aggregation, for instance, improves precision by 14%; combining machine learning with constraint satisfaction shows an improvement in recall by 23%.

Constraint-Based runtime prediction of SLA violations in service orchestrations

Service compositions put together loosely-coupled component services to perform more complex, higher level, or cross-organizational tasks in a platform-independent manner. Quality-of-Service (QoS) properties, such as execution time, availability, or cost, are critical for their usability, and permissible boundaries for their values are defined in Service Level Agreements (SLAs). We propose a method whereby constraints that model SLA conformance and violation are derived at any given point of the execution of a service composition. These constraints are generated using the structure of the composition and properties of the component services, which can be either known or empirically measured. Violation of these constraints means that the corresponding scenario is unfeasible, while satisfaction gives values for the constrained variables (start / end times for activities, or number of loop iterations) which make the scenario possible. These results can be used to perform optimized service matching or trigger preventive adaptation or healing.

Towards Data-Aware QoS-driven Adaptation for Service Orchestrations

Several activities in service oriented computing can benefit from knowing properties of a given service composition ahead of time. We will focus here on properties related to computational cost and resource usage, in a wide sense, as they can be linked to QoS characteristics. In order to attain more accuracy, we formulate computational cost / resource usage as functions on input data (or appropriate abstractions thereof) and show how these functions can be used to make more informed decisions when performing composition, proactive adaptation, and predictive monitoring. We present an approach to, on one hand, automatically synthesize these functions from orchestrations and, on the other hand, to effectively use them to increase the quality of non-trivial service-based systems with data-dependent behavior. We validate our approach by means of simulations with runtime selection of services and adaptation due to service failure.

Automatic Fragment Identification in Workflows Based on Sharing Analysis

In Service-Oriented Computing (SOC), fragmentation and merging of workflows are motivated by a number of concerns, among which we can cite design issues, performance, and privacy. Fragmentation emphasizes the application of design and runtime methods for clustering workflow activities into fragments and for checking the correctness of such fragment identification w.r.t. to some predefined policy. We present a fragment identification approach based on sharing analysis and we show how it can be applied to abstract workflow representations that may include descriptions of data operations, logical link dependencies based on logical formulas, and complex control flow constructs, such as loops and branches. Activities are assigned to fragments (to infer how these fragments are made up or to check their well-formedness) by interpreting the sharing information obtained from the analysis according to a set of predefined policy constraints.

An initial proposal for data-aware resource analysis of orchestrations with applications to predictive monitoring

Several activities in service oriented computing can benefit from knowing ahead of time future properties of a given service composition. In this paper we focus on how statically inferred computational cost functions on input data, which represent safe upper and lower bounds, can be used to predict some QoS-related values at runtime. In our approach, BPEL processes are translated into an intermediate language which is in turn converted into a logic program. Cost and resource analysis tools are applied to infer functions which, depending on the contents of some initial incoming message, return safe upper and lower bounds of some resource usage measure. Actual and predicted time characteristics are used to perform predictive monitoring. A validation is performed through simulation.

Building dynamic models of service compositions with simulation of provision resources

Efficient and competitive provision of service compositions depends both on the composition structure, and on planning and management of computational resources necessary for provision. Resource constraints on the service provider side have impact on the provision of composite services and can cause violations of predefined SLA criteria. We propose a methodology for modeling dynamic behavior of provider-side orchestration provision systems, based on the structure of orchestrations that are provided, their interaction, statistically estimated run-time parameters (such as running time) based on log traces, and the model of resources necessary for orchestration provision. We illustrate the application of our proposed methodology on a non-trivial real world example, and validate the approach using a simulation experiment.

A constraint-based approach to quality assurance in service choreographies

The knowledge about the quality characteristics (QoS) of service compositions is crucial for determining their usability and economic value; the quality of service compositions is usually regulated using Service Level Agreements (SLAs). While end-to-end SLAs are well suited for request-reply interactions, more complex, decentralized, multi-participant compositions (service choreographies) typically need multiple message exchanges between stateful parties and the corresponding SLAs thus involve several cooperating parties with interdependent QoS. The usual approaches to determining QoS ranges structurally (which are by construction easily composable) are not applicable in this scenario. Additionally, the intervening SLAs may depend on the exchanged data. We present an approach to data-aware QoS assurance in choreographies through the automatic derivation of composable QoS models from participant descriptions. Such models are based on a message typing system with size constraints and are derived using abstract interpretation. The models obtained have multiple uses including run-time prediction, adaptive participant selection, or design-time compliance checking. We also present an experimental evaluation and discuss the benefits of the proposed approach.

Automated Attribute Inference in Complex Service Workflows Based on Sharing Analysis

The properties of data and activities in business processes can be used to greatly facilitate several relevant tasks performed at design-and run-time, such as fragmentation, compliance checking, or top-down design. Business processes are often described using workflows, and we present an approach to mechanically infer business domain-specific attributes of workflow components, including data items, activities, and elements of sub-workflows, from known attributes of workflow inputs and the structure of the workflow by modeling these components as concepts and applying sharing analysis applied to a Horn clause representation of the workflow. The analysis is applicable to workflows featuring complex control and data dependencies, embedded control constructs, such as loops and branches, and embedded component services.

Towards QoS prediction based on composition structure analysis and probabilistic environment models

Complex software systems are usually built by composing numerous components, including external services. The quality of service (QoS) is essential for determining the usability of such systems, and depends both on the structure of the composition and on the QoS of its components. Since the QoS of each component is usually determined with uncertainty and varies from one invocation to another, the composite system also exhibits stochastic QoS behavior. We propose an approach for computing probability distributions of the composite system QoS attributes based on known probability distributions of the component QoS attributes and the composition structure. The approach is experimentally evaluated using a prototype analyzer tool and a real-world service-based example, by comparing the predicted probability distributions for the composition QoS with the actual distribution of QoS values from repeated actual executions.

Exploring the impact of inaccuracy and imprecision of QoS assumptions on proactive constraint-based QoS prediction for service orchestrations

Constraint-based Quality of Service (QoS) prediction is a method for predicting violations of Service Level Agreements (SLAs) in an executing instance of a service orchestration. It uses assumptions about the ranges of QoS values for component services in the orchestration. Experiments suggest that the method, when given correct component QoS assumptions, produces highly accurate predictions according to a series of quality-of-prediction metrics, and that it does so well ahead of the time when the prediction is to happen. We study the behavior of this method when the component QoS assumptions become incorrect or too vague. We conclude that the effect is a graceful deterioration in prediction quality, unless gross (order-of-magnitude) imprecisions are introduced. However, the method is very sensitive to the loss of information on the lower bounds for component QoS values, since the knowledge of the upper bounds is not sufficient for failure prediction.