HR: 1340h
AN: IN13B-1210 [Abstracts]
TI: Applications of Server Clustering Technology in Sensor Networks
AU: * Davis, G
EM: gadavis@ucsd.edu
AF: Scripps Institution of Oceanography, University of California San Diego, 9500 Gilman Drive,
La Jolla, CA 92093-0225, United States
AU: Foley, S
EM: sfoley@ucsd.edu
AF: Scripps Institution of Oceanography, University of California San Diego, 9500 Gilman Drive,
La Jolla, CA 92093-0225, United States
AU: Battistuz, B
EM: bbattistuz@ucsd.edu
AF: Scripps Institution of Oceanography, University of California San Diego, 9500 Gilman Drive,
La Jolla, CA 92093-0225, United States
AU: Eakins, J
EM: jeakins@ucsd.edu
AF: Scripps Institution of Oceanography, University of California San Diego, 9500 Gilman Drive,
La Jolla, CA 92093-0225, United States
AU: Vernon, F L
EM: flvernon@ucsd.edu
AF: Scripps Institution of Oceanography, University of California San Diego, 9500 Gilman Drive,
La Jolla, CA 92093-0225, United States
AU: Astiz, L
EM: lastiz@ucsd.edu
AF: Scripps Institution of Oceanography, University of California San Diego, 9500 Gilman Drive,
La Jolla, CA 92093-0225, United States
AB:
The Array Network Facility is charged with the acquisition and
processing of seismic data from the Earthscope USArray experiment.
High resolution data from 400 seismic sensors is streamed in near
real-time to the ANF at UCSD in La Jolla, CA where it is
automatically processed by machine and reviewed by analysts before
being externally distributed to other data centers, including the
IRIS Data Management Center. Data streams include six channels of 24-
bit seismic data at 40 samples per second and over twenty channels of
state-of-heath data at 1 sample per second per station. The sheer
volume of data acquired and processed overwhelms the capabilities of
any one affordable server system. Due to the relatively small buffers
on-site (typically four hours) at the seismic stations, it is vital
that the real-time systems remain online and acquiring data around
the clock in order to meet data distribution requirements in a timely
manner. Although the ANF does not have a 24x7x365 operations staff,
the logistical difficulty in retrieving data from often remote
locations after it expires from the on-site buffers requires the real-
time systems to automatically recover from server failures without
immediate operator intervention.
To accomplish these goals, the ANF has implemented a five node Sun
Solaris Cluster with acquisition and processing tasks shared by a
mixture of integer and floating point processing units (Sun T2000 and
V240/V245 systems). This configuration is an improvement over the
typical regional network data center for a number of reasons:
- By implementing a shared storage architecture, acquisition,
processing, and distribution can be split between multiple systems
working on the same data set, thus limiting the impact of a
particularly resource-intensive task on the acquisition system.
- The Solaris Cluster software monitors the health of the cluster nodes
and provides the ability automatically fail over processes from a
failed node to a healthy node.
- Redundant power and networking connections reduce the chances of a
single hardware failure taking down an entire node.
- By allowing processing systems to be transferred between cluster
nodes, Sun Cluster provides the ability to take servers down for both
planned and unplanned maintenance.
- This is not a "grid system", there is no single controlling node to
fail and take down the entire cluster.
UR: http://anf.ucsd.edu
DE: 0910 Data processing
DE: 0935 Seismic methods (3025, 7294)
DE: 7294 Seismic instruments and networks (0935, 3025)
SC: Earth and Space Science Informatics [IN]
MN: 2007 Fall Meeting