Blackboard

How can the task of compiling or constructing a dataset be divided up into smaller tasks?

Context

Applications consuming Linked Data commonly need to compile a dataset by collecting data from a number of distributed sources, e.g. other Linked Data datasets, SPARQL endpoints, or converting data from other legacy systems. A monolithic approach to constructing an aggregated dataset can be slow, fragile, and complex to implement. Dividing up tasks into smaller units of work can help parallelize the compilation of the data.

Solution

Create a number of separate processes that are each specialised to a particular task. Each process runs independently from each of the others, allowing each to be as simple or as complex as necessary. The processes all co-ordinate in the Progressive Enrichment of a single dataset, but are not triggered in any particular sequence. Instead each process looks for specific patterns in the underlying dataset, triggering the processing of the data, the results are then written back to the dataset.

Example(s)

An application is written to monitor twitter for uses of a particular hash tag. The application provides a simple RDF view of each tweet, including its author, text, timestamp. The RDF is written into a shared RDF dataset that serves as the Blackboard for a number of other processes. Additional processes are then written that query the dataset to discover new tweets. These processes carry out discrete tasks. The results of the tasks may trigger further processing by other tasks, or may be used to directly drive specific application behaviour. The tasks may include steps such as:

  • Looking for profile information about the author of each tweet
  • Attempting to find a FOAF profile for each author
  • Extracting hash tags from new tweets
  • Extracting URLs from individual tweets
  • Discovering metadata associated with URLs mentioned in tweets

Discussion

The Blackboard pattern is an existing design pattern that has been used in a number of existing systems, it works well when applied to the task of aggregating RDF and Linked Data due to the ease with which a dataset can be enriched over time.

The RDF dataset used as the "blackboard" shared by the independent processes may be short-lived, e.g. an in-memory datastore used to respond to a particular request or task, or permanent, e.g. an on-going aggregation of data on a particular topic or theme.

The decomposition of the data aggregation & conversion tasks into smaller units makes it easier to explore different approaches for implementing the desired behaviour, e.g. to explore alternate technologies or algorithms. The overall result of the processes co-operating to compile and enrich the dataset can be extremely complex but without requiring any overall co-ordination effort. Additional processes steps, e.g. to acquire data from additional sources, can easily be added without impacting on the overall system, making the architecture extensible.

Related