Information Technology Reference
In-Depth Information
maintains a variable “CPU Availability” which
indicates how much CPU power it can offer to run
peers' search tasks. When the node is busy with
other tasks or when the owner wants to start using
the node, CPU-Availability is set to zero.
query vectors is not obligatory, since one way to
speed up retrieval is to process only a fraction of
such query vectors.
A common method between peers for hashing
binary files to a signature value is assumed, so
that identical files must have identical signatures
and the chance of different files having identical
signatures to be very low.
The work considers both a replicated database
and a general P2P scenario, while special attention
is given on the control of the workload produced
at queried peers during query time. Each query
is divided into two phases, the first of which in-
cludes only a subpart of the actual query vectors,
in order to distinguish high probability response
peers. Accordingly, a peer ranking occurs and
the full query vectors are sent to all peers. Given
that a peer has free CPU resources, it decides
whether to process a query or not based on the
ranking that the specific query received, among
other factors.
The work proposed by C. Yang (2003), is based
on music indexing framework (MACSIS), which
reports high accuracy levels in most of the five
levels of musical similarity: (1) identical digital
copies; (2) same analog source, different digital
copies, possibly with noise; (3) same instrumental
performance, different vocal components; (4)
same score different performances, possibly at
different tempo; (5) same underlying melody,
different otherwise, with possible transposition)
used therein. As the MACSIS framework is
highly computationally intensive, its proposed
utilization in a P2P environment is safeguarded
by a the “CPU-availability” variable, allowing
peers to select their engagement to the retrieval
process. Moreover, the use of different sampling
levels of the query vectors as well as a two
staged search phase, in the later of which higher
probability response candidates may perform
more demanding tasks, allows a further level of
security in terms peer CPU-availability as well
as network utilization.
Feature Extraction
The music indexing framework (MACSIS) (C.
Yang, 2002) consists of three phases: (1) Raw
audio goes through a series of transformations,
including Fourier Transformation, leading to a
stream of characteristic sequences, which are
high-dimensional vectors representing a short
segment of music data. (2) Characteristic se-
quences for the audio database are then indexed
in a high-dimensional indexing scheme known
as Locality-Sensitive Hashing, or LSH (Indyk
& Motwani, 1998). (3) During retrieval, Phase 2
finds a list of matches on characteristic sequences,
representing short segments of music, which re-
quire joining together to determine which song
is the best “global” match.
Searching Phase
A music query issued by one node (querier)
in the network, is processed by a set of other
nodes, which may include the querier itself. Due
to the symmetric nature of P2P networks, any
node can become a querier and a node may be
simultaneously a querier and response node for
different queries. A query is sent as a stream of
characteristic sequences, obtained by analysing
user-input audio at the querier. Each query has
two required parameters: expiration-time and
search depth, indicating the time at which the
query expires, and the maximum number of
links through which the query can be passed in
the P2P network graph. The query is sent from
the querier to its neighbors, which may in turn
pass it on to other neighbors of their own (while
decrementing search-depth by 1) as long as search
depth is greater than zero. The processing of all
Search WWH ::
Custom Search