CISTI'2014 - 9ª Conferencia Ibérica de Sistemas y Tecnologías de Información

Full Program »

Improvements to Efficient Retrieval of Very Large Temporal Datasets with the TravelLight Method


A considerable number of domains deal with large and complex volumes of temporal data. The management of these volumes, from capture, storage, search, transfer, analysis and visualization, provides interesting challenges. One critical task is the efficient retrieval of data (raw data or intermediate results from analytic tools). Depending on the user's task, the data retrieved may be too large to fit in the local memory and, even when it fits, the time taken to retrieve the data may be too long. Previous work proposed the TravelLight method which reduced the turnaround time and improved interactive retrieval of data from large temporal datasets by exploring the temporal consistency of records in a database. The underlying main idea is that, at any given moment, a request is sent for data intersecting the temporal focus of interest which is not already marked as loaded in the database. To do this, the proxy module keeps track of loaded data at the database level, which generates time overheads from select and deselect operations. The current work propose an improved version that adopts a new paradigm focused in time intervals instead of solely in data items. This paradigm shift enables the separation of the proxy module from any particular temporal data source, as it is autonomous and efficient in the management of retrieved data. Our work also demonstrates that the overheads introduced by the new paradigm are smaller than prior overall overheads, further reducing the turnaround time. Results are reported concerning experiments with a temporally linear navigation across two datasets of one million items. With the obtained results it is possible to conclude that the approach presented in this work, built over the original TravelLight method further reduces turnaround time thus enhancing the response of interactive tasks over very large temporal datasets.

Author(s):

Alexandre Valle de Carvalho    
FEUP / INESC Porto
Portugal

Marco Amaro Oliveira    
INESC Porto / FEUP
Portugal

Artur Rocha    
INESC Porto / FEUP
Portugal

 

Powered by OpenConf®
Copyright ©2002-2013 Zakon Group LLC