Identifiers In Kepler

This is version 12. It is not the current version, and thus it cannot be edited.
[Back to current version]   [Restore this version]

To help scientists manage and share data sets, workflows, and workflow computation steps (i.e., actors) as well as semantic descriptions (i.e., metadata and ontology-based annotations) of each of these, we propose adding a new file-management middleware subsystem to Kepler.

The main goals of the file-management subsystem are listed below. We use the term "Kepler item" loosely below to refer to actors, workflows, and data sets, and "annotations" to refer to metadata (like EML) and semantic descriptions. The subsystem should provide the infrastructure for enabling scientists to:

• Search for all known Kepler items of interest.
• Organize Kepler items of interest using a file-directory metaphor (currently called "actor libraries" in Ptolemy). For example, a scientist should be able to create and persist a personal library, and to browse and search that library.
• Allow Kepler items to be organized into multiple libraries, and in multiple places within a library.
• Persist Kepler items in a network-accesible repository (i.e., in the EcoGrid).
• Retrieve new Kepler items from a network-accessible repository (the EcoGrid) and update changes to local items.
• Track both revisions of Kepler items as well as new versions (off-shoots or branches) of an item.
• Provide similar functionality for annotations, i.e., store annotations of actors, workflows, and data sets and publish those annotations to a network repository (the EcoGrid), and retrieve new annotations from the network repository.

The figure below outlines the architectural components of the subsystem. We treat annotations and Kepler items uniformly below. That is, the management subsystem does not have separate storage components for managing annotations and items. The subsystem assumes the use of Life-Sciences Identifiers (LSIDs) as logical identifiers for items and annotations. (QUESTION: and also for local "libraries" -- do these need to be stored in the EcoGrid, e.g.?). Thus, relevant files are assigned LSIDs and are accessed via LSIDs. We assume that an LSID will store the type of file (e.g., the type of annotation, or the type of Kepler item) associated with the ID.

Fig 1: A high-level architecture for file-management in Kepler.

The remote manager component provides the operations required to interact with the EcoGrid. It also may provide a local cache for better performance. The LSID manager creates and assigns unique identifiers for items and annotations. The logical/physical index manager provides operations to relate physical files to LSIDs, and to access physical files based on LSIDs. The directory/view manager provides operations to support the creation and retrieval of local, customized libraries.

The following are operations (these are very half-baked!) may be provided by the Kepler file-management subsystem:

• Retrieve(ID). Given an LSID, returns the relevant files associated with the item.
• LocalIDs(IDMetadata). Retrieve a list of local LSIDs, based on LSID metadata.
• RemoteIDs(IDMetadata). Retrieve a list of remote LSIDs, based on LSID metadata.
• Store(ItemHandle, IDMetadata). Construct an LSID with the given metadata for the item, and store the item in the local repository.
• GetRemoteUpdates(). Retrieve a list of new or updated LSIDs from a remote source.
• Update(LSID). Update the given LSID item from a remote source. This retrieves a new revision if one exists.
• Branch(LSID). Create a new version of an item. This creates and returns a new LSID for the item.
• CommitToRemoteServer(LSID). Update this item in a remote server.

MOTIVATION FOR LSIDs

To link metadata and semantic annotations to actors and workflows that are utilized in Kepler, we need a consistent scheme for uniquely identifiying these components. Currently, MoML refers to the implementing Java class as the principal definition of the actor, but this does not allow for the specializations that might occur later that constrain and define the actors I/O signatures and functionality. For example, the 'Expression' actor can be specialized by providing a particular expression to be evaluated, and the I/O signature of this specialized actor can be far more constrained than the Expression actor is generally.

In SEEK, we wish to provide both a structural and a semantic description of the signature and behavior of the actors and services used in models. This will allow us to use these descriptions to construct more powerful search and browsing services and to help integrate and compose workflows.

The EcoGrid and Taxon communities within SEEK are adopting Life Science Identifiers (LSIDs) as the principal syntax for creating identifiers. These ientifiers are free of semantics relating to the identified object, which makes it far easier to maintain consistent identifiers for a set of changing objects. LSIDs are described more thoroughly in EcoGridIdentifiers.

Proposal: Use LSID identifiers for actors, services, and components

Potential components needing identification

• An actor or service overall
• This would enable annotations regarding the behavior of the actor overall
• Would need to attach these identifiers to Java implementations of actors, to specializations of actors, and to web service descriptions (e.g., WSDL), among other things
• A port from an actor or service
• This enables annotation of the actor signature, both before and after specialization
• Combinations of ports? Probably not, as they can be referred as compound objects

References