NASA. Ames Research. Center. Moffett Field, California. 94035. Edward. Tejnil ... job. These include pre-proeessing input files, logging into the compute system.
AUTOMATED
CFD
PARAMETER
DISTRIBUTED
Stuart
E.
STUDIES
PARALLEL
Rogers,
Michael
COMPUTERS
Aftosmis,
and Center
NASA
Ames
Research
Moffett
Field,
California
Edward
Tejnil
and
ON
Shishir
Pandya
94035
Jasim
Ahmad
Eloret Moffett
Field,
California
94035
Introduction
As
Computational
Fluid
computational-resource run
large
costs
parameter
for a complete
Dynamics continue
studies.
aerospace
vehicle
depending
on
requirements
for viscous
CFD
advantage
wall-clock
time
are available comes trade
the
and
include there,
must
pre-proeessing executing
use of simple But when
a full-time
scripts
one wants
job,
is particularly
true
when
heterogeneous
compute
and most
thousands the
compute
platforms
of the
If enough
analysis control
databases.
attempting
to run
of mundane
tasks
into the compute
job, rather
tedious,
when
of these
tasks
of job,
a more
resources
at more
then
and
are
spread
out
the
the
this
including
study. CFD
than
With
ways,
parameter
more
or
platforms
study.
archiving
will help when sophisticated
to reduce
and copying
running
solvers
solution.
computing
for each
system
post-processing,
possible
of unique
a large
to
Computational
parameter
in a number
as
PC
of an Euler
parallel
a large
CFD
on a high-end
that
it becomes
and
methods
Euler
vehicle.
10 to 20 times
hour.
a number
the
to perform to run
and
files, logging
monitoring
complexity
matures,
to use CFD
of 5 to 20 hours
to perform
CFD
software
for inviscid
platforms, an
when
performing
input
and
This can become
arise
possible
the
possible
stability
of difficulties
it becomes
are typically
high-end
in building
and
order
computing
it becomes
time
and
to less than
of using
spend
solver
technologies
requirements
on the
solvers
solution
possibility
studies
the
are
of parallel
to a user,
A number a user
per
to drop,
Computational
workstation,
By taking
(CFD)
Often
job.
These
the input solver
files
output.
a few jobs.
The
running
a few dozen
jobs.
processes
is required.
This
over
a number
if different
one location. 1OF4
NASA
Information
Power
NASA
Computing,
Information
the
Computing,
software
and
objective
automated
the
the
need
the
use
for
possible.
process
user
and
user
of the
(IPG) and
Networking,
for security
The
Grid
monitoring
work
Such
a software
approach
taken
Systems and
jobs
computers system
and
been
The
software
single
a massive
developed,
tll_' under
IPG
provides
submission.
resources.
of every
under
Program,
Project.
a prototype
to populate has
job
focus
(CICT)
(CNIS)
on IPG
intervention
a major
Technologies
is to build
CFD
and
is currently
Communication
authentication,
of running
different
This
Information
current
of many
effort.
This
system
CFD
job.
run
and
system
matrix
should
will
remove
It should
in the
is known
which
enable
shortest
time
as the
AeroDB
script
several
discrete
mod-
system. The ules.
These
each
include
individual
rameter The
study.
following
AeroDB
and
The
details
section
on the
development
a database,
job,
for the
section
for the
of the
provides
lessons
was to build
module,
a run-manager
a job-launcher
a web-based
analysis
of AeroDB
user
design
the
results
AeroDB
a Remote AeroDB all the
Execution design
scripts
attributes. mand;
launch
in this effort,
The
remote the step
is shown
and
flow
database.
The
the
each job
in the database,
AeroDB
effort
by the
provided was
(GCS)
database.
also
CFD
and
included flow solvers
a mechanism
and
The
for the
jobs
used
for each
using
concludes
with
a
area.
for the
for re-running use of a Run
to monitor
convergence
flow-solver
to automatically
and
globusrun
run
user
Manager
until
library
execution the
monitored
converged
each
about
to see the restarting
the
for each
reporting
information
and
remaining
com-
from
and
while
for
and job
was executed
files,
an interface
and
the
files, executed
to enter
hub
on information
script
used
a mechanism
of the
information
using
output
Launcher,
communication
job based
the
a Job
A flowchart
all status
Execution
were
server,
as the
the input
scripts
development
pa-
section.
performed
in this
portal.
for execution
Remote
provided
a web
to store
post-staged
portal
provided the
and
pre-processed
and
was
work
of the
following
paper
on a MySql
was
resource
Broker.
web
script,
database
job-submission
The
service
it was used
compute
The
for future
progress
in the which
(RLV).
ideas
submitted
post-processed
jobs into
used
script
appropriate
and
1. The
presented
to monitor
Design
submission
interface;
job; it pre-staged
solver,
to the
web
Launcher
Services
compute
in Fig.
of the
study
vehicle
of a database
a job
the user
Job the
Common
consisted
script,
and
it chose
Grid
scripts
are
of a parameter
AeroDB The
for monitoring
of AeroDB
of a reusable
learned
portal
module
new
status
of
jobs.
The
which
was
time.
This
solution
obtained. 20F4
Fig.
1 AeroDB
flowchart.
Results
The
AeroDB
for a Liquid Cart3D hours
1000
We continued
13 different
shows and
globusrun
the
approximate resources
(LGBB)
had
cases
resources job
starting
number and
cases four
had
hours
function.
cases
Jobs 9:00
At
AM
over the
on
end
on each
time,
Many enough sent host.
No special-priority
of these
computing to each
and
Within
had
2863 These
study
Overflow
10th. cases
successfully.
job submissions
the
September
of this
locations.
parameter
both
100 Overflow
to obtain
used
for a large
to execute
completed
different
in order
of CPU
monitor
and
days.
of globusrun
their
at
completed
submissions number
used
at
and vehicle.
for seven
211 Overflow
compute
to execute
launched cases
the approximate
compute the
were
Cart3D
and
used
Booster
to execute
completed,
multiple
were
Glide-Back
flow solvers over
had
scripts
completed.
Cart3D cases cases time.
compute
Table
1 also
lists
queues
were
used
72
cases utilized
required Table
1
resource, the
other
on any
computers.
3OF4
of
Location
Host
Hardware
ARC
chapman.nas.nasa.gov
SGI
O3K
3489
25485
lomax.nas.nasa.gov
SGI
O3K
1074
15678
steger.nas.nasa.gov
SGI
O2K
477
8017
hopper.nas.nasa.gov
SGI
O2K
411
4702
evelyn.nas.nasa.gov
SGI
O2K
61
262
simak.nas.nasa.gov
Sun
Ultra
136
234
sharp.as.nren.nasa.gov
SGI
O2K
126
1014
aeroshark.as.nren.nasa.gov
Linux
PC
70
976
NCSA
modi4.ncsa.uiuc.edu
SGI
O2K
99
483
ISI
jupiter.isi.edu
SGI
O2K
GRC
# of Jobs
Hrs
21 5964
Total
CPU
212 57065
Conclusion
The tem, study,
final version
of the results and
of this paper
and
of suggestions
effectiveness for future
will include
more
of the current
details
approach
of the design for automating
of the
AeroDB
a CFD
sys-
parameter
improvements.
4OF4
