►
Description
Part of the Migrating from Cori to Perlmutter Training, December 1, 2022.
Please see https://www.nersc.gov/users/training/events/migrating-from-cori-to-perlmutter-training-dec2022/ for the training day agenda and presentation slides.
A
Hello:
everyone,
my
name,
is
Helen
hee
and
user
engagement
group.
Welcome
to
today's
introduction
to
migrating
from
Quarry
to
promoter
training.
A
A
few
Logistics
first,
everyone
is
muted
and
also
please
change
your
name
in
Zoom
to
first
name
last
name
and
your
username.
You
can
do
click
on
participants
and
more
next
to
your
name
to
rename
closed
caption
is
enabled,
so
you
can
click
the
CC
button
on
and
off
and
also
an
option
to
view
for
transcript.
You
can
also
save
transcripts
if
you
wish
and
will
provide
slides
after
each
talk,
we
will
also
process
videos
and
make
them
available
after
in
a
few
days.
A
Our
staff
are
are
available
standing
by
and
can
bring
you
to
breakout
rooms.
Maybe
they
can
answer
questions
in
the
Q.
A
also
I
want
to
mention
that
this
is
like,
if
you're
already
doing
lots
of
you
know
coding
and
running
jobs
on
from
a
GPU.
This
may
not
be
the
training
for
you,
it
may
be.
You
may
think
this
is
too,
you
know
basic.
A
So
it's
sort
of
geared
towards
people
are
starting
migrating
from
their
query
applications
to
to
promoter,
but
you
are
welcome
to
stay
in
or
skip
some
talks
and
come
back
for
the
Hands-On
exercises
and
because
slides
and
talks
will
be
available
afterwards.
So
you're
welcome
to
you
know,
review
them
as
you
need
so
to
to
be
more
efficient
for
the
time.
A
A
The
allocation
2023
will
be
based
on
Farmers
capability,
and
the
nurse
hours
allocated
can
be
used
on
query
on
those
allocations
from
a
CPU
and
permanent
GPU.
So
the
permanent
CPU
allocation
is
shared
with
the
usage
on
query,
so
we
would
like
to
give
users
more
time
to
transition
your
your
applications.
So
as
of
now,
we
haven't
decided
official
date,
the
capital
T
in
as
in
March
2023
and
we'll
be
using
the
T
in
next
slide
to
show
some
to
share
some
of
the
retirement
timeline.
A
So
software
has
already
been
frozen,
which
means
no
new
user-facing
software
will
be
installed
by
nurse
and
from
now
November
to
January
2020
to
January.
We
are.
We
have
scheduled
lots
of
office
hours
and
next
one
is
tomorrow
we
will
come
to
to
join
and
final
date.
T
will
be
announced,
late,
January
or
early
February,
and
then
one
week
before
the
retirement
date.
A
How
to
access
promoter
you
would
do
SSH
your
username
at
promoter.p1.,
nurse.gov
or
your
username
at
saw
p1.net.gov.
So
is
the
first
name
of
parameter
just
for
easy
typing
and
you
again
would
use
MFA
with
multi-factor
authentication,
which
is
password
plus
one
time
password
the
same
way
as
you
do
on
query.
You
could
also
use
SSH
proxy
to
reduce
the
frequency
of
authentication.
The
default
time
is
24
hours,
so
you
don't
have
to
type
password
again
and
again,
if
you
logging
multiple
times
during
the
day,
I
also
used.
A
Many
people
also
use
the
Jupiter
Hub
to
access
promoter.
So
this
is
what
you
can
see
and
we
have
a
formula
you
have
shared
CPU,
node
and
exclusive
CPU
and
GPU
nodes,
there's
also
configurable
GPU,
which
allows
to
to
say
using
reservations
and
having
more
adjustable
parameters
that
you
can
set.
There's
a
terminal
kernel
of
it.
A
So
as
long
as
as
well
as
you
can
choose
many
other
Jupiter
kernels
in
Jupiter
Hub,
if
you
choose
open
the
terminal,
then
you
have
a
terminal
and
either
on
any
of
those
nodes
that
the
terminal
deck
you
can
edit
and
see
your
file.
A
Just
to
quickly
mention
the
file
system
on
data
considerations,
so
for
for
Quarry
you
have
Global
Home
and
Community
file
system
directories.
Those
are
all
available
on
parameter,
except
there's
a
Sim
length
that
users
are
getting
used
to
Curry,
which
is
global,
slash
project,
slash
projectors.
A
It
is
this
links
to
community
file
system
and
this
Sim
link
will
not
be
ported
to
Perimeter.
So
if
your
scripts
are
still
using
that
be
sure
to
update
to
use
the
new
the
real
Community
file
system
directories,
one
thing
to
point
out
is
that
Corey
scratch
is
not
accessible
on
poor
mother
format
has
its
own
p-scratch
and
current
scratch
will
be
retired
with
Corey.
So,
in
order
to
migrate
Curry
scratch
data
onto
Community
devices
onto
perimeter,
either
to
communicate
a
system
or
Global
home,
you
can
use
the
Globus
or
SCP.
A
If
you
want
those
data
to
be
available
on
promoter
scratch,
there's
another
step
that
you
need
to
move
from
from
the
files
on
CFS
or
if
hpss
you
can
there's
a
Globus
endpoint
for
parameter
that
you
can
move
data
over
the
link
has
all
the
details:
okay,
a
few
similarities
for
Corey
and
formatter
comparisons.
The
user
environment
is
pretty
similar.
We
also
use
prg
EnV
modules
and
the
compiler
wrappers
to
to
build
applications.
You'll
see
here
more
in
the
next
CPU
talk,
and
we
also
use
the
slum
parameter
and
has
very
similar
cues.
A
Regular
premium
overrun,
shared
Etc,
especially
on
CPU
shared,
is
only
available
on
CPU
and
the
compute
nodes
on
the
CPU
compute
nodes,
as
AMD
instead
of
Intel,
but
with
a
very
similar
CPU
architecture
on
human
domain.
Number
of
numinum
is
different,
but
very
similar,
architectures,
no
major
surprises
and
has
similar
to
halfway
clock,
speed
and
similar
in
Kenya.
As
regard
to
number
of
cores
per
node.
A
Some
major
differences
on
formula
we
use
lmod
Quarry
use
tickle,
that's
still
pretty
similar,
but
has
some
differences,
one
of
them
like
module,
available
versus
module,
spider
and
modular
with
versus
module.
Swap
and
you'll
hear
more
in
the
CPU
talk
as
well.
A
They
may
have
different
versions
for
GPU,
compatible,
binaries
or
CPU
code
application,
CPU
versions,
so
using
GPU
is
the
topic
of
today's
training
just
to
point
out
we're
not
training
on
the
program.
The
program
models
are
for
GPU,
we
have
our
head
and
the
last
we
have
I
have
a
slide
showing
all
the
previous
trainings
on
using
different
program
models
of
MP,
open,
ECC
code
Etc.
A
So
training
today
is
more
on
practical
usage
on
using
GPU
nodes,
compiler
differences,
there's
no
Intel
compiler,
which
is
the
default
on
query,
and
we
currently
do
not
have
a
plan
to
support
it
and
the
the
default
program.
Environment
per
model
is
canoe
and
we
have
other
Nvidia
and
Clay
compilers
available.
A
So
as
I
mentioned,
this
is
a
few
existing
training
materials
on
using
poor
mother
on
programming
models
Etc
and
the
the
pink
line
shows
all
the
training
events.
You
can
find
the
2022
events
there
and
then
some
links
over
archives.
Then
you
find
archives
that
2021
or
the
previous
years
and
have
we
have
slides
and
recordings
available
for
majority
of
the
trainings
for
most,
if
not
all,
and
feel
free
to
explore
using
parameter
training.
A
We
was
focused
on
GPU
in
January,
and
new
user
training
is
also
geared
towards
using
parameter
GPU
for
science
day
and
lots
of
programming
model
and
user
applications.
Their
stories
are
porting
applications
to
promoter
day-to-day
is
focused
more
on
data
usage,
data,
workflows
and
tools,
and
then
we
have
openmp
of
load
code,
open,
SEC,
Circle
and
the
Kodi
is
a
development
tool
that
helps
you
to
inserting
or
openmp
and
open
ACC
directives
for
your
applications
in
video
hpcsdk
is
all
about
the
Nvidia
Suite,
with
supporting
openmp,
open,
SEC
could
and
also
standard
parallelization.
A
Or
upcoming
training
opportunities,
as
I
mentioned
office
hours,
we
have
held
three
in
November
and
talked
to
over
50
users,
additional
office
hours
already
scheduled
for
December
and
January,
and
when
upcoming
one
is
tomorrow
and
another
Kodi
training
is
under
under
planning,
then
also
we're
also
planning
a
no
way
for
GPU
programming
boot
camp
of
offered
by
Nvidia,
open
ACC.
So
we'll
be
talking
about
these
programming
models
and
having
a
mini
challenge
in
the
boot
camp.