Grid Resource Allocation - Grid Computing: Infrastructure, Service, and Applications - page 166

7. 2 .1 High-Throughput Computing

For many experimental scientists, scientii c progress and quality of
research are strongly linked to computing throughput. In other words,
most scientists are concerned with how many l oating-point operations
per month or per year they can extract from their computing environment
rather than the number of such operations the environment can provide
them with per second or minute. High-throughput computing (HTC) [1]
environments focus on large amounts of processing capacity over long
periods of time.
The key to HTC is effective management and exploitation of all avail-
able computing resources. The main challenge in a grid environment is
how to handle resource characteristics of autonomy, heterogeneity, and
geographic distribution and maximize the amount of resources accessible
to an application.
Large-scale parameter sweep [2] is a classic scenario requiring HTC.
A parameter-sweep application model is a combination of task and data
parallel models, and applications formulated to use this model contain a
large number of independent jobs operating on different datasets. A range
of scenarios and parameters to be explored are applied to program input
values to generate different datasets. The programming and execution
model of such applications resembles the Single Program, Multiple Data
(SPMD) technique. The execution model essentially involves processing N
independent jobs (each with the same task specii cation, but a different
dataset) on M distributed computers where N is, typically, much larger
than M . Fortunately, this high-throughput parametric computing model
is simple, yet powerful enough to formulate distributed execution of many
application areas such as radiation equipment calibration analysis, search-
ing for extra-terrestrial intelligence, protein folding, molecular modeling
for drug design, human-genome sequence analysis, brain activity analysis,
high-energy physics events analysis, ad hoc network simulation, crash
simulation, tomography, i nancial modeling, and Mcell simulations.

7. 2 . 2 Resource Co-Allocation

Resource co-allocation in a grid refers to the simultaneous assignment to
application multiple resources belonging to possibly different admini-
strative domains [ 3 ] . Basically, two types of applications need resource
co-allocation: (1) computationally intensive applications that may require
more resources than a single computing machine can provide, and
(2) applications requiring several types of computing capabilities from
resource owners in other administrative domains. For example, a scien-
tii c instrument, i ve computers, and multiple display devices were used
for collaborative real-time reconstruction of X-ray source data [4]. While
such simultaneous allocation can in principle be achieved manually,