2. Why
this
Experiment
?
Access
to
remote
computing
resources
Focus
on:
simulation
applications
in
digital
manufacturing
and
High
Performance
Technical
Computing
(HPTC)
Focus
on:
remote
resources
in
HPTC
Centers
&
in
HPTC
Clouds
Observation:
while
business
clouds
are
becoming
widely
used,
the
acceptance
of
simulation
clouds
in
industry
is
still
in
an
early
adopter
stage
(CAE,
Bio,
Finance,
Oil
&
Gas,
DCC)
Reason:
big
difference
between
business
and
simulation
clouds
Barriers:
Complexity,
IP,
data
transfer,
software
licenses,
parallel
communications,
specific
system
requirements,
data
security,
interoperability,
cost,
etc.
3. Our
Goal
for
the
Experiment
Current
Experiment:
August
–
October
2012
Form
a
community
around
the
benefits
of
HPTC
in
the
cloud
Hands-‐on
exploring
and
understand
the
challenges
with
digital
manufacturing
in
the
Cloud
Study
each
end-‐to-‐end
process
and
find
ways
to
overcome
these
challenges
Document
our
findings
4. Participants
Some
of
our
Providers
Some
of
our
Resource
Providers
want
to
be
anonymous
Media
Sponsor
5. Participants
Some
of
our
ISVs
want
Some
of
our
Software
Providers
to
be
anonymous
6. Participants
Some
of
our
HPC
Experts
Some
of
our
HPC
Experts
want
to
be
anonymous
7. Participants
Many
of
our
industry
end-‐users
Some
of
our
End-‐Users
want
to
be
anonymous
8. Where
are
we
with
the
experiment
We
currently
have
over
170
participating
organizations
and
individuals
Experiment
reaches
to
every
corner
of
the
globe,
participants
are
coming
from
22
countries
Participants
sign
up
through
www.hpcexperiment.com
and
www.cfrdexperiment.com
25
teams
have
been
formed
and
are
active
9. Participants
by
geography
%
of
Site
Traffic
US
36
%
Germany
12
%
Italy
6
%
Australia
6
%
Spain
5
%
UK
5
%
Russia
3
%
France
3
%
Other
24
%
10. Teams,
it’s
all
about
teams
Anchor
Bolt
Cement
Flow
Wind
Turbines
Resonance
Sprinkler
Combustion
Radiofrequency
Space
Capsule
Blood
Flow
Supersonic
Car
Acoustics
ChinaCFD
Liquid-‐Gas
Dosimetry
Gas
Bubbles
Wing-‐Flow
Weathermen
Side
impact
Ship-‐Hull
ColombiaBio
11. Building
the
teams
An
end-‐user
joins
the
experiment
Organizers
(Burak,
Wolfgang)
identify
perfect
team
expert
Organizers
contact
the
ISV
and
ask
to
join
the
experiment
End-‐user
and
team
expert
analyze
resource
requirements
and
send
to
organizers
Organizers
suggest
one
or
two
computational
resource
providers
After
all
four
team
members
agree
on
the
right
resource
provider,
the
team
is
ready
to
go
12. Bumps
on
the
road
–
the
top
4
Delays
because
of
vacation
times
in
August
&
other
projects
(internal,
customer)
from
our
participants
Getting
HPC
participants
was
quite
easy,
getting
CAE
participants
was
a
challenge
Participants
can
spend
only
small
portion
of
their
time
Learning
the
access
and
usage
processes
of
our
software
and
compute
resource
providers
can
take
many
days
Process
automation
capabilities
of
providers
vary
greatly.
Some
have
focused
on
enrollment,
registration
automation,
while
others
haven’t.
Experiment
organizers’
lack
of
automation,
currently
the
whole
end-‐to-‐
end
process
is
manual
(intentionally)
Getting
regular
updates
from
Team
Experts
is
a
challenge
because
this
is
not
their
day
job
Consider:
the
sample
size
is
still
small
13. Are
we
discovering
hurdles?
Reaching
end-‐users
who
are
ready
and
willing
to
engage
in
HPTC
and
especially
HPTC
in
the
Cloud.
About
half
of
our
participants
want
to
remain
anonymous,
for
different
reasons
(failure,
policies,
internal
processes,…)
HPC
is
complex.
At
times
it
requires
multiple
experts.
Matching
end-‐user
projects
with
the
appropriate
resource
providers
is
tricky
and
critical
to
the
teams
success.
Resource
providers
(e.g.
HPC
Centers)
often
face
internal
policy
and
legal
hurdles
Sometimes,
the
1000
cpu-‐core
hours
are
a
limit
14. Let’s
hear
from
Team
Experts!
Chris
Dagdigian
Co-‐founder
and
Principal
Consultant
BioTeam
Inc
Ingo
Seipp
Science
+
Computing
15. Team
2
Short
Status
Report
HPC
Expert:
End
User:
Anonymous
16. Team
2
Overview
OUR
END
USER
OUR
HPC
SOFTWARE
• Individual
&
organization
has
requested
CST
Studio
Suite
anonymity
“Electromagnetic
Simulation”
• Goal:
Hybrid
model
in
which
local
and
www.cst.com
cloud
resources
leveraged
simultaneously
Diverse
Architecture
Options
• We
can
say
this:
1. Local
Windows
Workstation
– It’s
a
medical
device
2. CST
Distributed
Computing
– Simulating
new
probe
design
for
a
particular
device
3. CST
MPI
– Tests
involve
running
at
simulation
size
&
resolution
that
cannot
be
performed
OS
Diversity
internally*
Various
combinations
of
Windows
– *Using
fake
data
at
this
time
and
Linux
based
machines
18. First
Design
Failed
Miserably
The
Good
The
Bad
• Looked
pretty
on
paper!
Can’t
launch
GPU
nodes
from
• Total
isolation
of
systems
via
inside
a
VPC
Amazon
VPC
and
custom
subnets
…
so
we
ran
them
on
“regular”
• VPC
allows
for
“elastic”
NIC
EC2
devices
w/
persistent
MAC
addresses.
– Awesome
for
license
servers
…
and
this
did
not
work
well
NAT
translation
between
EC2
• VPN
Server
allowed
end-‐user
and
VPC
private
IP
addresses
remote
resources
to
directly
join
our
cloud
environment
wreaked
havoc
with
CST
Distributed
Computing
Master
20. Second
Design
–
Good
So
Far
The
Good
The
Bad
• It
works;
we
are
running
tasks
Lost
the
persistent
MAC
across
multiple
GPU
solver
nodes
address
when
we
left
the
right
now
VPC;
need
to
treat
our
license
• Security
surprisingly
good
despite
server
very
carefully
losing
VPC
isolation
– EC2
Security
Groups
block
us
from
Unclear
today
how
we
will
the
rest
of
Internet
&
AWS
attempt
to
incorporate
remote
solver
&
workstation
• CST
License
server
now
running
on
resources
at
end-‐user
site
much
cheaper
Linux
instance
We
know
from
attempt
#1
that
CST
and
NAT
translation
• Clear
path
to
elasticity
and
high-‐ don’t
work
well
together
…
scale
simulation
runs
21. Next
Steps
Run
at
large
scale
Refine
our
architecture
We
might
move
the
License
Server
back
into
a
VPC
in
order
to
leverage
the
significant
benefit
of
persistent
MAC
addresses
and
elastic
NIC
devices
Figure
out
how
bring
in
the
remote
resources
sitting
at
end
user
site
30. Team
14,
Short
Status
Report
Simulation
of
electromagnetic
radiation
and
dosimetry
in
Ingo
Seipp
humans
in
cars
induced
by
and
Team
mobile
phone
technology
Science
+
Computing
35. Announcing:
Uber-‐Cloud
Experiment
Round
2
Why
‘Uber-‐Cloud’:
HPC/CAE/BIO
in
the
Cloud
is
only
one
part
of
the
Experiment,
in
addition
we
provide
access
to
HPC
Centers
and
other
resources.
Round
1
is
proof
of
concept
=>
YES,
remote
access
to
HPC
resources
works,
and,
there
is
real
interest
and
need!
Round
2
will
be
more
professional
=>
more
tools
instead
of
hands-‐on,
more
teams,
more
applications
beyond
CAE,
a
list
of
professional
services,
measuring
the
effort,
how
much
would
it
cost,
etc.
Existing
Round
1
Teams
are
encouraged
in
Round
2
to
use
other
resources
or
can
participate
in
forming
new
Teams.
Oct
15:
Call
for
Participation;
Nov
15:
Start
of
Experiment
Round
2
36. What
is
next?
07/24/2012
Publish
updated
kick
off
document
07/24/2012
Request
for
detailed
participant
profiles
08/10/2012
End-‐user
projects
submitted
08/17/2012
Resources
are
assigned,
end-‐user
projects
start
09/14/2012
Half
Time
meeting
webinar
10/15/2012
End-‐user
projects
are
completed
10/31/2012
Experiment
is
completed
11/15/2012
Experiment
findings
are
published
11/15/2012
Start
of
Experiment
Round
2,
Kick-‐off
at
SC
in
Salt
Lake
City
37. Conclusion
Response
to
the
Uber-‐Cloud
Experiment
is
overwhelming
Everybody
is
learning
and
working
along
their
very
specific
business
interest
At
least
20
of
the
25
teams
will
finish
successfully
and
in
time
97%
of
current
participants
will
continue
in
Round
2
Univa
Grid
Engine
Community:
forming
teams
which
explore
bursting
into
a
public
HPC
Cloud
from
their
Univa
GE
cluster
The
experiment
could
help
the
Grid
Engine
customers
be
more:
business
flexible,
profitable,
cost
effective,
customer
friendly…