Affiliations: School of Computer Science, Cardiff University, UK | DEIS University of Calabria, Rende (CS), Italy | CNR-ICAR, Rende (CS), Italy | Center for Computation and Technology, Louisiana State
University, USA
Note: [] Corresponding author: Eddie Al-Shakarchi, Cardiff School of
Computer Science, Cardiff University, Queen's Buildings, 5 The Parade, Roath,
Cardiff CF24 3AA, UK. Tel.: +44 (0)29 2087 4812; Fax: +44 (0)29 2087 4598;
E-mail: [email protected]
Abstract: This paper discusses issues in the distribution of bundled workflows
across ubiquitous peer-to-peer networks for the application of music
information retrieval. The underlying motivation for this work is provided by
the DART project, which aims to develop a novel music recommendation system by
gathering statistical data using collaborative filtering techniques and the
analysis of the audio itsel, in order to create a reliable and comprehensive
database of the music that people own and which they listen to. To achieve
this, the DART scientists creating the algorithms need the ability to
distribute the Triana workflows they create, representing the analysis to be
performed, across the network on a regular basis (perhaps even daily) in order
to update the network as a whole with new workflows to be executed for the
analysis. DART uses a similar approach to BOINC but differs in that the workers
receive input data in the form of a bundled Triana workflow, which is executed
in order to process any MP3 files that they own on their machine. Once
analysed, the results are returned to DART's distributed database that collects
and aggregates the resulting information. DART employs the use of package
repositories to decentralise the distribution of such workflow bundles and this
approach is validated in this paper through simulations that show that suitable
scalability is maintained through the system as the number of participants
increases. The results clearly illustrate the effectiveness of the
approach.
