Information Technology Reference
In-Depth Information
Figure 3. Effects of inexact runtime estimation under normal distribution
MOLDABLE JOB ALLOCATION IN
HETEROGENEOUS GRID
computing speed of a single processor. Similarly,
the required computing power of a job is defined
as the number of required processors specified in
the job multiplied by the computing speed of a
single processor on its home site.
In the following, we compare the perfor-
mances of five different cases. They are indepen-
dent clusters representing a non-Grid architecture,
moldable job allocation without runtime estima-
tion, moldable job allocation with exact runtime
estimation, moldable job allocation with uniform
distribution of runtime-estimation errors, mold-
able job allocation with normal distribution of
runtime-estimation errors. Figure 5 presents the
results of simulations for a heterogeneous com-
putational Grid with speed vector (1,3,5,7,9) and
load vector (10,10,10,10,10), where
In a computational Grid environment, a common
practice is try to allocate an entire parallel job onto
a single participating site. Sometimes a parallel
job, upon its submission, cannot fit in any single
site due to the occupation of some processors
by running jobs. How the job scheduler handles
such situations is an important issue which has
the potential to further improve the utilization
of Grid resources as well as the performance of
parallel jobs. This section extends the moldable
job allocation policies proposed in the previous
sections to deal with the resource fragmentation
issue in a heterogeneous computational Grid
environment.
The detailed moldable job allocation procedure
is illustrated in Figure 4. The major difference
between the moldable job allocation procedures
for a homogeneous parallel computer and for
a heterogeneous Grid environment is the site
selection process regarding the computation and
comparison of computing power of different sites.
A site's free computing power is defined as the
number of free processors on it multiplied by the
•
IC. Independent clusters.
•
no estimation. Adaptive processor alloca-
tion without runtime estimation.
•
exact estimation. Adaptive processor allo-
cation with exact runtime estimation.
•
uniform distribution. Adaptive processor
allocation with uniform distribution of
runtime-estimation errors.
Search WWH ::
Custom Search