Information Technology Reference
In-Depth Information
the use of the DART system would be a music
recommendation system, whereby the audio
(typically in mp3 format) on a user's machine is
analyzed and suitable recommendations are made
to the user based on the results of the analysis
carried out. The analysis of the files could be
index-based, including statistical correlations of
song names to extract commonalities in order to
make a recommendation— or the analysis could
be content based, actually searching for track
tempos, timbre, pitch, mood, frequency range,
and so forth. Algorithms for statistical and au-
dio-based analysis consist of workflows that are
created by the so-called
DART Manager
. Such
workflows are bundled into a
package
(containing
the Java code and their required resources) that is
propagated onto the network in order to upload
the new analysis to the users' machines. As the
algorithms are updated and refined by the DART
Manager, the updated Triana task graph and any
new tools are bundled into a new package, and
distributed onto the network.
The scenario described here discusses the
decentralized DART network, in which nodes
are organized in a super-peer topology using
the P2PS middleware. In P2PS producers (e.g.,
providers of packages containing algorithms or
results) create
adverts
to advertise that they have
something that is of interest to other participants
in the network. Consumers (i.e., the peers that
wish to use available packages, result sets and
so on) issue a
query
in order to search for rel-
evant adverts. P2PS
rendezvous
nodes are then
responsible for matching queries, with adverts
within their local cache, in order to search for
matches and respond accordingly. Consumers
receive adverts when their query matches, and
can be used to retrieve the relevant information
they require that is, download the new package to
perform the analysis. The
DART Manager
node
produces and advertises the bundle representing
the new DART package (called
DART Package
Adverts
) containing algorithms that the worker
nodes need to run (new Triana units and task
graphs). With the DART system, the data files
that undergo analysis (the music files and mp3s)
are on the users local system, and therefore all
data processing is local so network bandwidth is
not consumed transferring large data files over
the DART infrastructure. Although local, the
processing is massively parallel, as participants
analyze their own audio files in parallel.
Simple peers, or
workers
, are available for
package execution, and therefore issue a
pack-
age query
to download a package in order to
start their analysis. Super-peer interconnections
are exploited to make package queries travel
the network rapidly; super peers play the role of
rendezvous
nodes
, since they can store package
adverts (and potentially the packages themselves
if the super peers are also functioning as
Package
Repositories
, as discussed shortly) and compare
these files with queries issued to discover them;
thereby acting as a meeting place for both package
providers and consumers. Since packages could
require a reasonable amount of storage memory,
it is assumed that only some of the peers in the
network will cache such files. These peers are
referred to as
Package Repositories
(
PR
) nodes
and can be located on super peers or worker peers.
Each user or node in the system would decide
if they want to be a super peer and/or package
repositories, as well as a worker.
Figure 5 shows a sample topology with 5 super-
peers (2 of which are also package repositories),
and the sequence of messages exchanged among
workers, super-peers and package repositories to
perform the package submission protocol. These
messages are related to the execution of a packager
by a single worker, labelled as
W0
. Note that in
this example, normal peers are not considered as
package repositories.
The protocol requires that job execution be
preceded by two matching phases that exploit the
features of the super-peer network: the
package-
advertisement
phase and the
results-query
phase.
In the
package-advertisement
phase, the
DART
manager
generates a number of
package adverts
,
Search WWH ::
Custom Search