Model-Driven Analysis of Workflow Adaptations

Thesis supervisor: Mireille Blay-Fornarino (UNS/MODALIS)

Several production infrastructures exist to support various medical research needs (Health-e-Child project, NeuGrid project…). The development of Grid-based medical applications involves (i) defining a medical data analysis workflow (business procedure) while (ii) taking into account some non-functional requirements such as the need to conform to strict legal constraints in terms of data privacy and security, to support the large amount of data manipulated and to reliably support highly stressing amounts of computations.
In order to address the key non-functional requirements, the user has to design workflows in their contextual environment, thus spending time on non-business oriented tasks. Even if infrastructures are improving in terms of size, amount of data manipulated and amount of computations, the final user still has to ensure that the workflows conform to initial requirements and to cleverly decide when to duplicate some tasks, how to package set of tasks, when to re-submit faulty jobs, …
The workflows must be adapted according to legal changes, trade agreements, and context modifications. Some of these adaptations are operated statically, when designing new workflows by increment on existing ones. Other adaptations must be realized dynamically at runtime to react to faults or decrease in the quality of service. Due to different points of view - cost, efficiency, and security - several adaptations can occur on a same workflow. The user has again to ensure that the resulting workflow is still “valid”.

Based on these observations, there is a need for helping users to define accurate workflows and to adapt them in a safe way according to changes at design time and runtime.
To build such an adaptation environment, it is necessary to define a meta-model to express constraints on workflow and another meta-model to specify workflow adaptation supporting composition. These meta-models should capture the workflow in term of controls and data flows.

The objective of this thesis is to develop a framework for controlling workflow adaptations. This work encompasses i) a meta-model to capture workflow adaptations supporting incremental composition, ii) a meta-model to specify requirements on workflows and adaptations, iii) the detection at design time of inconsistencies and interactions between adaptations, iv) the detection at deployment time of inconsistencies and interactions between adaptations applied on a given workflow, v) the detection at runtime of inconsistencies due to adaptations of workflow (this step will be based on abstraction of workflow and its set of adaptations at runtime).

Agenda

The work to realize during this thesis will be organized as follows:

  • Study applications already developed to establish a set of requirements and adaptations to support scalability, systems faults management, security constraints and their interactions.
  • Study the state-of-the-art in the domain of workflow description languages (data or control flow definition languages, engines of workflow, Architecture definition languages), workflow analysis (Properties definition language, process algebra, constraint solving), workflow adaptation (aspects, distribution policies, business patterns). The study will particularly focus on the scalability and incremental issues of the related works in these domains.
  • Propose a workflow control meta-model to express constraints on workflows such as the one underlying legal and ethical requirements. A set of predefined models corresponding to given constraints will first be established.
  • Define a workflow adaptation meta-model for composing in an incremental and safe way adaptations such as the ones resulting from the state-of-the art studies. This meta-model should support in an incremental way (i) at design time, validation of one adaptation according to a set of constraints: may this adaptation interact with any of these constraints? (ii) at deployment time, validation of adaptations according to a set of requirements applying on a given workflow : Does the composition of these adaptations violate any requirement ? (ii) at run-time, a priori checking that an adaptation will not violate pre-established constraints. This last step assumes that we are able to keep consistent a model of the running application conforming to our meta-model.
  • Build a composition environment (if possible a plug-in to an environment dedicated to workflow management) that presents a user friendly interface to access several requirements and adaptations, to define new ones, to check for the consistencies of adaptations and requirements and to manage adaptations according to requirement consistency point of view.
  • Validate the comprehensive approach (i) by redefining existing applications using the developed tools and comparing the results, and (ii) by applying it on the application on image-based research on the Alzheimer Disease.