Traceability is an important feature of work ow based systems, and is a key source of provenance data. This paper presents CRISTAL, a mature software platform developed and used at CERN for experiment construction at the LHC. It is entirely work ow based capturing provenance on every aspect of its use from application development to end-user interaction. In this paper we summarize some initial work towards the adaptation of CRISTAL to a more semantic orientation, in particular compliance with the Open Provenance Model.
Provenance, as in the documentation of the origin or source of something [
The current CRISTAL model is already capable of capturing the
information required by the N4U project. However, the core of CRISTAL was developed
more than ten years ago, before the semantic web, and so although the data is
captured tools are missing to easily communicate that information to other
systems. Consequently, we have decided to modernise it by using emerging semantic
web based provenance models. This paper presents initial work that has been
undertaken at CERN to export the current CRISTAL provenance model to be
compliant with semantic provenance vocabularies such as the Open Provenance
Model (OPM) [
This paper is structured as follows: section 2 presents the CRISTAL platform and how it has been used previously, especially in the context of provenance. Section 3 presents the current CRISTAL provenance model and section 4 describes a preliminary attempt to convert it to the OPM. Finally section 5 presents conclusions with directions for future work. 2
CRISTAL is a system designed to provide data management, work ow tracking
and change management to an agreed set of user requirements. It is a distributed
data and work ow management system which uses a database for its repository,
a multi-layered architecture for its component abstraction and dynamic object
modelling for the design of its objects and components. These techniques were
deemed critical in handling the complexity of data-intensive systems and to
provide the exibility to adapt to the changing production scenarios typical
of any research production system. CRISTAL has been based on a so-called
description-driven approach in which all logic and data structures are described
by meta-data, which can be modi ed and versioned online as the description
of the object, component, item or an application changes. A Description-Driven
System (DDS) architecture, as advocated previously in [
The meta-data along with the instantiated elements of data are stored in the database and the evolution of the design is tracked by versioning the changes in the meta-data over time. Thus DDSs make use of meta-objects to store domainspeci c system descriptions that control and manage the lifecycles of domain objects. The separation of descriptions from their instances allows speci cation and management to evolve independently and asynchronously. This separation is essential in handling the complexity issues facing many web-based computing applications and facilitates interoperability, reusability and system evolution. Separating descriptions from their instantiation allows new versions of de ned objects (and in turn their descriptions) to coexist with older versions.
Neuroimaging is constantly developing new algorithms and work ows; these
may require variations to the provenance data that is collected. At the same
time provenance data, to be useful, needs to remain consistent over time, to be
traceable, to be queryable and easily accessible and scientists analyses need to
be conducted on those data. CRISTAL handles all of this. The reader is directed
to previous publications on DDS for further background ([
The current CRISTAL model is shown in gure 2. A collection of all these objects is known as an Item in CRISTAL terminology. An Item contains : { Work ows i.e. complete layouts of every action that can be performed on that item, connected in a directed graph that enforces the execution order of the constituent activities. { Activities capture the parameters of each atomic execution step, de ning what data is to be supplied and by whom. The execution is performed by agents. { Agents are either human users or mechanical/ computational agents (via an
API), which then generate events. { Events detail each change of state of an Activity. Completion events generate data, stored as outcomes. From the generation of an Event provenance information is stored. { Outcomes are XML documents resulting from each execution (i.e. the data from completion Events), for which viewpoints arise. { Viewpoints refer to particular versions of an Item's Outcome (e.g. the latest version or, in the case of descriptions, a particular version number). { Properties are name/value pairs that name and type items. Properties also denormalize collected data for more e cient querying, and { Collections enable items to be linked to each other.
The provenance is captured by Events being generated. For an Agent to write anything within CRISTAL they need to generate an Event. This is done by altering a state in an Activity. The actual provenance information required is application dependent. CRISTAL allows application designers to de ne their own backend with respect to how they wish to store provenance. Thus a key function of the CRISTAL system is its ability to adapt to changing requirements in terms of provenance storage. The domain of e-science is constantly changing as new work ows, algorithms and research studies are developed. The underlying CRISTAL model allows the system to evolve to handle such challenges whilst retaining provenance information in a consistent and traceable manner. For example in the neuGRID project the CRISTAL provenance service captured: { Work ow speci cations - These were XML based speci cations of work ow descriptions which were external to CRISTAL. These were serialised and stored in an relational database. { Data or inputs supplied to each work ow component - The parameters to each work ow component in the case of neuGrid and N4U these are images, however, they can be any piece of data. { Annotations added to the work ow and individual work ow components These consisted of simple name value pairs which allowed uses to store extra information about a work ow. { Links and dependencies between work ow components - This is a part of the work ow speci cation. { Execution errors generated during analysis - A necessary component for any provenance model. { Output produced by the work ow and each work ow component - In the case of neuGRID and N4U these are images, however, as wit inputs they can be any piece of data that the application requires.
these were all added onto CRISTAL as a relational database model. In the neuGRID project CRISTAL was exposed as a provenance service where information is captured in real time and stored in the current CRISTAL provenance database. However, to enhance CRISTAL we have decided to apply semantic web technologies to the current CRISTAL provenance model. The next section explains some early work on converting the CRISTAL model to be more compliant with the OPM. 4
The OPM is currently a provenance model which is gaining popularity and implementations are becoming available in OWL, RDF and Java. We personally feel that the OPM is a good choice for modernizing the current CRISTAL provenance model since the mapping for the \top layer" of the graphical model ts in well with the current CRISTAL provenance model. This top layer includes Artifact (a physical state such as the result of an action performed), Process (an action that is caused by an artifact and may generate a new artifact) and Agent (an entity which causes the execution of a process). Figure 3 shows the mapping of the tentative N4U CRISTAL Provenance Model onto the OPM.
The colours map onto the conceptual ideas of artifacts (e.g. A1, A2) and processes (e.g. P1) in the OPM. Agents map onto CRISTAL agents since they initiate Processes. Processes from the OPM are equivalent to Work ows and Activities from CRISTAL since they generate Artifacts. Artifacts from the OPM are CRISTAL Events and Outcomes because they are various forms of data. At the time that this initial mapping between the OPM and the CRISTAL provenance model was conceived, there was not an implementation of the OPM available. We are currently in the process of converting the CRISTAL provenance model into the OWL version of the OPM implementation. We believe that once this conversion has been completed it will allow us to demonstrate further how Semantic Web technologies can aid in the provenance of work ows. 5
In this research in progress paper we presented CRISTAL which is a system that is able to capture provenance information based on work ows. CRISTAL currently is being used in many di erent projects and is in the process of being commercialised. An initial mapping of the provenance aspect of CRISTAL to the OPM was shown.
Future work consists of expanding our initial mapping of the OPM to CRISTAL
from a simple and preliminary model to using the OWL implementation of the
OPM. We believe that this will aid compatibility with other work ow based
systems such as Taverna [