Database Reference
In-Depth Information
When a user submits a CQ, it is optimized and started in the client manager.
When the client manager identifies an SP, the subquery of that SP is registered
with the coordinator of the cluster where the subquery is to be executed
( feCC, bgCC, or beCC in Figure 11.10). Then, the coordinator starts an RP
to execute the subquery. In addition, an RP can dynamically start new RPs
by requesting them from the coordinator of the cluster where the new RP is
started.
11.4 Streaming Function Approximation
for Scientific Simulations
Simulations are one of the most important tools of modern science for studying
real world phenomena. The physical laws that govern a phenomenon drive a
mathematical model, usually a function defined on a multidimensional space,
that is used in simulations as an approximation of reality. In practice scientists
face serious computational challenges, because realistic models are expensive
to evaluate and simulation runs consist of a large number of model evaluation
steps. To make realistic simulations feasible, scientists are often willing to
accept approximate results as long as significant improvements in runtime
can be achieved. The main idea is to use previously computed function values
to construct an approximate model that can be used for future steps instead
of the expensive original model.
Speeding up simulations is challenging, because it inherently is a data stream
processing problem. Like data stream management systems, a simulation en-
gine has to process a stream of data points that describe the current state of
the simulated system. Decisions about which previously computed function
values to retain and how to leverage them for reducing the number of future
function evaluations have to be made in real time and with limited knowledge
about future data points. This again is a typical characteristic of data stream
applications.
However, despite fitting into the general class of data stream applications,
scientific simulations also have unique challenges. These create a novel data
stream indexing and data stream model learning problem. Techniques for
addressing these challenges are described in the remainder of this chapter.
Example
Consider the simulation of a combustion process, which motivated the line of
work discussed in this section. Scientists study how the composition of gases
in a combustion chamber changes over time due to chemical reactions. The
composition of a gas particle is described by a high-dimensional vector, typi-
cally with 10-70 dimensions. The simulation consists of a series of timesteps.
Search WWH ::




Custom Search