This topic was presented by Alexandru Arion at the 54th annual conference IEEE Global Communications Conference (GLOBECOM 2011) from 5 â 9 December 2011 in Houston, Texas.
Publication: http://bit.ly/A3iKbv
Abstract:
The trend for more online linked data becomes stronger. Foreseeing a future where "everything" will be online and linked, we ask the critical question; what is next? We envision that managing, query- ing and storing large amounts of links and data is far from yet another query processing task. We highlight two distinct and promising research directions towards managing and making sense of linked data. We in- troduce linked views to help focusing on specific link and data instances and linked history to help observe how links and data change over time.
2. Data networks
Local: low power connected devices transmit to base stations.
Large scale: base stations transmit over large distances using existing
communication infrastructure.
3. Relevance
Large numbers of sensor networks are already being
interconnected and share huge amount of streaming data.
Example: SwissEx (http://www.swiss-experiment.ch)
4. Related work
S. Shah, et all., âAn efficient and resilient approach to filtering and disseminating
streaming data,â in VLDB, 2003, pp. 57â68.
Y. Zhou, et all., âDisseminating streaming data in a dynamic environment: an
adaptive and cost-based approach,â The VLDB Journal, vol. 17, no. 6, pp. 1465â
1483, 2008.
D. J. Abadi, et all., âThe design of the Borealis stream processing engine,â in
CIDR, 2005, pp. 277â289.
M. Balazinska, et all., âLoad management and high availability in the medusa
distributed stream processing system,â in SIGMOD, 2004, pp. 929â930.
P. Pietzuch, et all., âNetwork-aware operator placement for stream-
processing systems,â in ICDE, 2006, p. 49.
6. Key features (1)
Feature 1: reduces communication costs (does not
require any data transfer of actual streams)
Feature 2: any type of queries can be processed (all
data required for query processing is available to
consumer nodes)
7. Key features (2)
Feature 3: any type of model can be employed
(serves any application)
Feature 4: systematic solution that can guarantee
user-specified accuracy requirements for model-
based views.
8. Algorithms (1)
Coded model update:
â predetermines parameter values
â encodes them with bitmaps
â updates models efficiently sending only bitmaps
10. Framework properties
Accuracy requirements solution:
â The producer node generates a model-driven value when a new raw
reading is streamed, and checks whether the difference between the
raw value and the model-driven value stays within the error bound.
11. Â
â If the difference does not exceed the error bound, no communication is
required between the two nodes, and the consumer node generates
values for their model-based views.
12. Â
â Otherwise, the producer node reconstructs its model, so that the
model-driven value generated from the reconstructed model does not
exceed the error bound from the current raw reading. Next, the
producer node updates the models at consumer nodes by sending
new parameter values of the reconstructed model.
19. Further related work
A. Deshpande and S. Madden, âMauveDB: supporting model-based user views in
database systems,â in SIGMOD, 2006
Y. Ahmad, O. Papaemmanouil, U. Cž etintemel, and J. Rogers, âSimultaneous
equation systems for query processing on continuous-time data streams,â in ICDE,
2008
A. Thiagarajan and S. Madden, âQuerying continuous functions in a database
system,â in SIGMOD, 2008
A. Deligiannakis, Y. Kotidis, and N. Roussopoulos, "Compressing historical
information in sensor networks,â in SIGMOD, 2004
H. Chen, J. Li, and P. Mohapatra, âRACE: time series compression with
rate adaptivity and error bound for sensor networks,â 2004
S. Gandhi, S. Nath, S. Suri, and J. Liu, âGamps: Compressing multi
sensor data by grouping and amplitude scaling,â in SIGMOD, 2009