►
From YouTube: NUG Monthly Meeting Aug 2022 - ERCAP
Description
Recording of the NUG Monthly meeting for Aug 2022. Our topic of the day is the ERCAP process for requesting a NERSC allocation for AY2023
A
So
we'll
fold
basically
the
same
pattern
as
we
typically
do
reminder
this
is
this
is
interactive.
This
is
not
just
presenting
to
users.
This
is
a
discussion,
so
yeah
please
participate.
So
we've
got
a
couple
of
dozen
people.
I
think
that's
enough
that
that's
a
suitable
size
that
yeah
feel
free
to
just
unmute
and
and
speak
up.
If
it
starts
getting
too
noisy,
then
we'll
get
a
hands
up
or
whatever,
but
but
yeah
by
all
means
join
the
chat
and
I
see,
will
has
a
hand
up
already.
B
So
I'd
like
to
look
for
beta
testers
for
hpc
toolkit
on
perlmutter,
I've
built
it
last
week
and
it
works
with
pc
sampling
as
well,
but
without
papi.
So
I'm
still
looking
for
better
testers
to
see
whether
it
works
for
all
applications
or
not
things
like
that.
So
shoot
me
up
chat
on
the
slack
if
you
are
interested,
so
I
can
tell
you
where
the
module
is.
A
Cool
do
do
that,
but
we'll
come
back
around
to
that.
A
few
points
standing
in
the
agenda
when
we'll
actually
have
a
series
of
announcements
and
and
cfps,
including
this
sort
of
thing
where
yeah
there's
not
just
announcements
from
nurse,
gets
announcements
across
across
nag,
so
we'll
go
through
our
kind
of
normal
order
win
of
the
month
today
I
learned
a
series
of
announcements
and
our
topic
of
the
day
is
okay.
You
will
have
seen
that
the
okay
process
for
ay
2023
is
now
open.
A
So,
to
start
out
with
win
of
the
month,
so
the
yeah.
The
idea
here
is
to
show
off
an
achievement
or
shout
out
somebody
else's
achievement
that
you
are
aware
of
and
yeah
it
can
be
big
or
small,
from
soul
to
bug
that
was
yeah,
giving
you
trouble
for
a
while
through
to
having
a
paper
accepted
and
yeah
some
of
these.
Actually
one
of
our
announcements
coming
up
is
going
to
be
to
do
with
scientific
achievement
and
your
hpc
awards.
A
B
B
That's
correct
and
with
pc
sampling
you
can
get
into
the
kernel
and
correlate
like
blind
information
from
inside
the
gpu
kernel
to
say
you
have
data
from
stalls
or
things
like
that
inside
the
hardware
registers
I'm
surprised
as
well.
It
works
because
paramater
hasn't
been
patched
with
the.
What
was
it
called
again:
the
vulnerability
with
hardware
counters
issue
that
was
announced
a
couple
months
back.
A
A
Something
you
will
probably
find
is
so
on
quarter
infillmonte.
We,
we
use
a
system
called
ldmx
which
stands
for
lightweight
data
metric
system
to
to
collect
data
about.
You
know
how
kind
of
how
the
nodes
are
running,
and
you
know
one
of
the
the
sort
of
the
aims
with
that
is
to
be
able
to
pull
out
performance
information,
the
catches
of
course.
That
means
that
it's
doing
counter
sampling
as
well.
A
So,
if
you're,
using
tools,
probably
probably
including
hpc
toolkit,
there's
a
option
in
here
in
we
have
some
notes
on
it.
Now
in
our
documents,
you
need
to
do
a
bit
of
search
for
it,
but
yeah
you
essentially
need
to
to
switch
off
the
ongoing
counter
collection
just
for
your
job,
so
that
you
can
get.
B
A
C
Sorry
could
you
just
to
follow
up
I'm
not
too
familiar
with
hpc
toolkit.
I
just
found
from
the
google,
but
not
quite
sure.
If
this
is
the
correct
one,
can
you
put
the
link
that
to
the
hpc
toolkit
that
was
just
talked
about?
I'm
curious
if
you
could
use
that
on
part
matter
to
check
how
what's
the
bottleneck
of
my
code,
I'm
using
oh
there's,
one
sure.
A
A
C
So
we
just
had
a
second
virtual
meeting
among
members
and
I
am
working
on
adding
documentation
to
ask
on
how
to
really
compile
and
run
wolf
and
then
some
best
practices.
You
need
to
work
to
be
shared.
So
that's
actually
my
window
balance.
Now.
Finally,
I've
been
you
know,
procrastinating.
How
to
edit
document
ask
documentation,
then.
Finally,
I
forced
myself
to
learn
and
do
that.
C
The
one
question
is
that
at
first
I
thought
it's
good
to
put
this
new
web
page
under
applications,
but
looks
like
this
page.
Applications
are
those
available
already
from
mask
as
module
so
not
really
applicable
to
the
wolf
model
we
are
using.
So
it's
it's
freely
available
code
you
have
to
download
and
then
compile
not
really
available.
C
The
module,
so
I'm
looking
for
really
where
it
is
a
good
place
to
put
that
kind
of
information,
and
in
this
user
group
style,
I
think
we
will
find,
is
quite
pretty
informative,
useful
for
other
users
who
does
have
faced
similar.
You
know
difficulties.
You
know
that
this
particular
publicly
available
code
is
being
used
for
the
research,
but
it's
maybe
frequently
updated
or
not,
systems
also
being
updated,
and
many
users
particularly
graduate
students
having
sometimes
difficulties
to
compile
and
run
so
maybe
sometime
we
once
we
get
this
going.
C
A
So
so,
first
up,
congratulations
and
thank
you
for
contributing
to
our
documentation.
There's
yeah,
I
think,
there's
a
lot
of
benefit
in
getting
input
from
from
users
who
are
using
the
system
for,
as
you
discover
things,
things
that
work
and
things
that
aren't
that
trivial.
In
terms
of
that
application
section,
the
intent
of
that
section,
like
my
view
of
the
intent
of
that
section,
is
it's
to
help
users
to
use
applications
on
nurse
systems.
A
D
C
E
A
And-
and
so
it's
also
a
good
place
to
actually
ask
a
question
like:
is
this
the
right
place
for
it?
And
there
might
be
some
discussion
about?
Oh,
maybe
it
would
fit
better
here
or
you
know,
maybe
more
people
would
see
it
here
or
and
and
kind
of,
or
what
you
thought.
You
need
the
mode
request
into
into
something
that
fits
nicely
and
yeah.
A
A
Let's
move
on
to
we
could
we
can
still
talk
about
wins,
but
the
let's
also
talk
about
the
the
flip
side
of
that
coin,
which
is
today,
I
learned.
Sometimes
when
you
don't
have
a
win,
there's
actually
a
lot
of
benefit
in
it,
because
you
discover
what
doesn't
work
and
yeah
and
in
the
process,
sort
of
learn
more
about
how
you
know
how
things
work
at
a
deeper
level,
which
leads
to
wins
down
the
track.
D
Well,
I'll
I'll
pipe
in
I
run
a
code
called
ascot
which
simulates
the
orbits
of
charged
particles
and
fusion-grade
plasmas.
It's
written
by
a
group.
In
finland
it
was
written
to
be
optically
sort
of
optimized
for
a
kno
like
architecture
and
like
many
users,
I've
been
working
to
try
to
port
or
to
build
a
version
of
this
on
on
pearl
mudder,
and
I
was
surprised
that
it
rans
more
slowly
on
pearl
mudder
than
on
knl
by
about
20
to
30
percent,
and
I
did
a
fair
amount
of
documentation.
D
We
have
had
conversations
with
both
support
expert
experts
at
nurse
and
the
developer,
and
the
developer
claims
that
he
has
built
ascot
on
an
amd
like
or
an
amd
ship,
but
an
earlier
version.
That's
not
milan
and
he
claims
to
get
a
factor
of
two
increase
in
performance,
which
is
that
that
surprised
me
an
intel
compiler
on
amd,
so
yeah.
So
of
course
it's
a
statement.
I
would
also
like
to
know
whether
any
anybody
else
has
observed
you
know
not
getting
a
big
improvement
going
from
quarry
to
perlmutter
on
on
cpu-bound
tasks.
C
F
Hey
steve,
this
is
richard
at
nurse.
That's
interesting,
I
don't
know
other
steve
lee
if
you've
heard
of
reports
like
this,
I
I
do
know
that
I
mean
it
sounds
like
there's
something
just
wrong
in
a
configuration
or
something
I
know
we
did.
We
did
look
at
a
number
of
cpu
applications
and
codes
that
we
had
and
tried
them
out
on
on
milan
on
the
on
the
node
and
the
average.
F
F
D
D
A
So
something
that
I
have
seen
a
couple
of
times-
and
I
know
like
hpa
engineers-
are
working
on
improving
is
there
are
there
are
some
circumstances
where,
with
I
think
it's
essentially,
the
the
communication
between
processes
with
a
lot
of
processes
per
node
can
in
some
circumstances,
basically
be
a
bit
slower
and
there's
there's
some
kind
of
knobs
that
we're
still
working
to
tweak
on
that.
So
I
wonder,
if
that's
a
factor,
are
you
running
at
a
fairly
high
mpi
account?
A
A
A
A
If
I
recall
correctly,
for
today,
a
bunch
of
them
are
in
the
weekly
email,
of
course,
the
yeah,
the
big
one
being
that
the
economic
allocations
season
is
upon
us
and
you'll
have
a
you
know
a
much
deeper
dive
into
that
fairly
shortly.
You
will
have
seen
announcements
and
possibly
noticed
some
effects
of
cfs
and
peace
scratch
up
grades,
so
the
cfs
upgrade
update
should
now
be
minimally
disruptive.
I
think
the
engineers
are
still
working
on
things,
but
the
part
that
kind
of
you
know
directly
impacts.
A
Another
big
one
that
you
probably
saw
is
that
cory's
retirement
is
now
in
planning.
So
it's
intended
to
be
what
yeah
the
plan
is
for
corey
to
retire
at
the
end
of
this
allocation
yeah,
which
is
which
will
be
sort
of
about
early
on
january
january
17th,
if
I
remember
rightly
so
after
that.
Also
all
of
next
year's
allocations
will
be
on
perlmatter,
which
is
all
the
more
reason
to
get
codes
ready
for
pearlmatter.
A
On
perlmutter,
over
the
last
sort
of
several
weeks,
you
probably
notice
announcements
about
we're,
updating
the
commander's
gpu
nodes
to
use
slingshot
11,
which
is
the
I
guess,
the
newer
version
of
its
interconnect.
You
know
it
involved
a
hardware
update
of
the
nodes
as
well,
which
is
why
those
were
being
sort
of
taken
out
and
updated
and
put
back
in
so
that
is
now
complete
and
all
of
pearl
mata
is
slingshot,
11.
A
sure
simply
use,
and
you
should
see
some
benefits
other
big
one.
I
do
think
this
was
in
the
weekly
email,
but
you
hopefully
saw
it
the
other
day
in
a
direct
email.
Formatted
charging
will
begin
so
that
is
yeah
charging
for
perlmutterjobs
will
begin.
After
that
p
scratch
update
is
completed
so
yeah,
beginning
of
september.
A
I
think
there's
there's
more
detail
about
that
in
the
email,
hopefully
we'll
help
with
controlling
the
queue.
There's
quite
a
lot
of
demand
at
the
moment-
and
you
know
the
queues
are
getting
a
little
bit
long
yeah.
So
we
encourage
you
to
get
look
at
what
jobs
you're
submitting
and
I,
I
guess,
prioritize
with
the
the
knowledge
that
charging
will
start
soon.
A
A
One
is
for
high
impact
scientific
achievement,
so
you
know
something
that
is
high.
Impact
is
is
important
to
the
world
and
the
other
is
for
innovative
use
of
high
performance
computing.
So
if
you're
exploring
new
new
ways
to
use
or
new
new
things
to
do
with
hpc
we're
very
interested
and
there's
awards
available
for
these-
and
I
think
richard
can
correct
me
on
this-
I
think
there's
two
sort
of
categories-
one
for
early
career
and
one
sort
of
open
category
is
that
true.
A
A
Do
we
need
to
provide
scaling
test
data
on
palmata
for
just
filling
out
proposed
allocations?
I
think
I'll
defer
that
question
to
the
the
topic
of
the
day
while
clayton
is
talking
so
we'll
come
back
to
that
on
file
system
licenses,
we're
using
a
quarry,
pulmonary
scratch
might
not
be
available
for
selection
or
not
documented.
A
A
A
A
A
A
A
E
Okay,
so
my
name
is
clayton
bagwell,
I'm
with
the
nurse
account
and
allocation
support
in
the
business
operations
and
support
group
and,
as
you've
heard,
oops
get
on
the
right
screen.
Here
there
we
go
ercap
is
the
energy
research
computing
allocations
process?
Many
of
you
who
are
pis
know
that
this
is
how
you
request
access
to
resources
at
nursk.
E
The
aircap
application
interface
is
at
ercap.nurse.gov
and
can
be
accessed
by
anyone.
Don't
have
to
be
a
pi
using
your
nurse
username
password
and
mfa
one-time
password
ercap.
We
is
used
to
renew
current
projects
for
the
coming
year
and
also
to
submit
requests
for
new
projects.
You
have
new
research
grant
and
you
want
to
start
a
project
for
ay2023.
E
E
The
aircap
requests
are
reviewed
by
and
the
doe
office
the
science
program
managers
allocation
managers
and
they
provide
the
awards
of
computing
time,
requests
that
have
passed
in
the
process
and
will
be
awarded
they'll
get
an
award
email
sent
to
them.
In
december
and
this
year
the
er
in
2023
the
allocation
year
starts
on
january
18th.
E
E
So
the
currency
of
computing
time
is
note
hours
which
most
of
you
are
aware
of.
Some
of
the
older,
I
won't
say
older.
The
more
people
have
been
around
that
nurse
for
a
while
know
that
we
used
to
use
what
was
called
a
nurse
hour
just
a
way
of
trying
to
calibrate
how
we
charge
for
time
now
that
we
have
pro
mutter
we're
using
perlmutter
as
our
baseline
and
allocations
and
charging
are
now
in
node
hours.
E
This
year,
the
available
cpu
note
hours
is
about
15.8
million
hours,
and
this
is
a
little
chart
showing
how
the
time
is
distributed
amongst
the
various
doe
office
of
science
programs
and
there's
a
link
to
a
plate,
a
page
that
explains
it
a
little
more
gpu
time.
This
year
is
about
seven
and
a
half
million
gpu
note
hours
and.
E
E
E
E
Once
you've
started
a
request,
you
don't
have
to
finish
it
in
one
sitting.
You
can
save
a
draft
and
come
back
and
finish
it
later,
and
those
will
end
up
in
the
section
for
draft
requests
once
you've
submitted
your
request
and
submitted
it
and
once
you've
finished
your
request
and
submitted
it.
I
got
it
it'll
end
up
under
the
submitted
requests
down
here
and
for
continuing
projects.
There'll
be
a
section
for
previous
requests.
E
E
E
Once
you
get
into
the
aircap
request,
we
do
try
to
give
you
a
lot
of
help
text
using
highlighting
highlighting
information
in
color
boxes.
The
ones
in
pink
are
will
help.
You
we'll
highlight
things
that
are
mandatory,
so
if
this
is
a
renewal
request,
you
want
to
put
in
the
number
from
the
previous
request
that
you
want
to
renew
in
this
box
and
that
will,
like
I
said,
help
pre-fill
some
of
the
information
in
the
request
from
your
previous
request.
C
E
E
E
So
there
are
mandatory
items
that
will
be
highlighted
with
asterisks
red
means
that
you
need
there's
a
field
that
you
do
need
to
fill
in
and
also
we
have
many
of
our
questions
divided
up
under
these
tabs
under
the
top
section
here.
If
you
see
an
asterisk
next
to
the
label
in
the
tab,
that
means
that
there's
a
question
that's
mandatory
under
that
tab.
E
And
some
of
the
fields
will
have
these
magnifying
glasses
next
to
them,
which
means
that
they're
list
options
that
you
can
select
from
there's
another
way
to
search
through
those
lists.
One
is
to
type
in
like
two
asterisks
that
will
give
you
a
longer
display
of
the
options.
This
one
has
this
one.
It
shows
21
options,
but
it
actually
turns
out.
There's
22.,
so
there's
other
ways
to
refine
your
your
search,
you
can
type
in
a
keyword
and
that
will
shorten
the
list
to
the
items
that
have
that
keyword
in
them.
E
E
E
The
next
tab
is
for
funding.
Most
of
most
aircap
projects
are
are
funded
by
the
doe
office
of
science.
You
know
when
you
click
on
the
check
box
next
to
the
office
of
science.
Here,
you'll
have
additional
fields
show
up
asking
for
more
information
such
as
the
program
office.
That's
funding
your
research,
the
name
of
the
program
manager
that
approved
your
your
grant
and
then
underneath
this
list.
There's
a
box
for
the
actual
grant,
number
information
or
pams
or
fpw
information.
E
E
If
your
funding
doesn't
come
from
the
doe
office
of
science
or
u.s
agency
or
ldrd
funding,
then
we
do
request.
We
do
require
you
to
submit
information
on
how
your
project's
research
is
relevant
to
the
particular
doe
office
of
science.
That
you're
submitting
this
request
to
your
research
needs
to
be
in
alignment
with
the
program
mission
and
if
it's
not
you're,
less
likely
to
receive
an
award.
E
Next
is
a
security
tab,
so
nurse
supports
only
open
research,
that's
intended
to
be
published
in
scientific
journals,
etc.
We
don't
allow
proprietary
research
and
particularly
the
following
areas
such
as
classified
or
controlled,
military
defense,
information,
export
or
itar
personally,
research
in
uses
personally
identifiable
information
or
protected
health
information.
E
E
E
E
This
is
a
short
kind
of
a
low
level
or
what's
called
high
level,
some
scientific
american
level
of
information
about
what
your,
what
your
research
is
about
and
what
you
intend
to
accomplish
this.
The
second
section
is
going
to
be
a
more
detailed
description,
which
is
what
the
doe
managers
are
going
to
focus
on,
for
what
you're
trying
to
accomplish
what
your
research
goals
are
and
the
process
etc.
E
These
two
sections
do
have
fairly
large
character
limits,
but
if
you
get
something
more
than
like
3
000
characters
in
your
explanation,
you
probably
want
to
submit
or
attach
a
document
that
gives
the
the
more
in
detail.
Information
to
your
request
and
a
little
later
on,
we'll
show
you
about
how
to
attach
other
documents
to
your
request.
E
If
you're
renewing
your
project,
which
most
people
will
be
doing,
there's
additional
sections
below
that
that
are
required
your
accomplishments
from
using
nurse
over
the
past
year
and
also
any
publications
that
you've
had
published
or
scheduled,
be
published
from
using
nurse
resources.
What
we're
really
focusing
on.
If,
if
nothing
else,
if
you
can
provide
the
doi
for
your
document,
we
can
always
get
the
other
additional
information
to
fill
in
into
our
database.
E
E
E
E
E
For
storage,
hps,
archival
storage
and
the
cfs
project
storage,
these
amounts
are
requested
in
terabytes,
and
you
do
need
to
request
at
least
enough
storage
to
cover
what
your
current
usage
is.
If
not
you'll
be
given
a
warning,
information
and
you'll
need
to
re-enter
a
higher
number
and
again
you
we
don't.
We
only
accept
integers,
so
anything
with
a
decimal
or
fraction,
whatever
will
be
truncated,.
E
Let
us
know
how
you
determine
how
many
node
hours
you
need
for
next
year
and
how
much
storage
you
need
just
give
us
your
back
of
the
envelope
calculations
for
how
you
figured
out
how
much
time
you
need
we'll
also
ask
for
information
about
any
key
events
or
deadlines.
Basically
we're
looking
to
see
how
consistent
usage
will
be
throughout
the
year,
and
this
helps
us
determine
you
know
the
highs
and
lows
of
when
people
will
be.
E
E
E
E
E
And
once
you're
finished
filling
out
your
form
and
if
you
want
a
hard
copy
of
what
you've
filled
out,
there's
a
button
at
the
top
of
the
form
to
create
a
pdf.
The
system
will
create
the
pdf
and
automatically
attach
it
to
the
air
cap
request
and
it'll
also
show
up
down
here
under
the
attachment
section.
E
A
A
I
think
there
are
a
couple
of
questions
in
in
the
chat
I
can
run
through
them.
The
first
one
actually
was
from
before
you
started.
So
in
that
there
was,
there
was
some
comments
about
gpu
readiness
bin
asks.
May
I
ask
if
we
need
to
provide
scaling
test
data
on
perlmatter
for
justifying
our
allocation.
E
A
E
A
E
Less
than
that,
3
000
characters
or
less
that's
what
the
fields
will
hold.
So
if,
if
you
feel
that
your
explanations
or
I
mean
a
summary
shouldn't,
be
more
than
a
page
right,
the
yeah
more
detailed
descriptions
can
be
longer
but,
like
I
said,
3
000
characters
if
it's
anything
longer
than
that,
it's
better
to
just
attach
a
a
longer
document
to
the
ercap
request
and
it
all
gets
combined
together
and
sent
to
the
doe.
E
A
C
Oh
yeah,
thanks
do
you
still
have
time
great
time
sure
yeah
thanks?
I
have
couple
questions
so
I'm
preparing
allocation
application
for
not
research,
but
the
the
user
group
like
special
interest
group
at
nas,
I
kind
of
mentioned
earlier
during
the
meeting.
So
it's
a
user
group
among
us
users.
The
exchange
information
about
this
particular
called
wolf.
That's
advocacy
model
because
we
do
want
to
we.
We
do.
C
We
want
to
provide
documentations
like
how
to
compile
how
to
run
the
best
practices
to
run
this
model
and
then
to
do
that.
We'd
like
to
have
small
allocation
of
computing
hours
in
storage
and
we'd
like
to
have
that
as
soon
as
we
can.
So
we
are
preparing
applications
for
the
year
2022
this
year,
but
then
2023
is
opening
up.
So
I'm
wondering
if
I
can
submit
the
applications
to
both
yes
at
the
same
time,
or
should
I
apply
for
the
current
year,
wait
get
approved
and
then
renew
all
right.
E
C
E
C
E
Well,
for
some
very
large
projects,
they
they're
already
scheduled
to
to
run
for
multiple
years,
like
it's
a
like
a
five-year
plan
or
something,
and
so
even
though
their
cap
request
is
only
geared
for
one
year
at
a
time.
It'll
help
us
know.
If
you
plan
to
be
using
nurse
for
the
next
five
years,
we
can
help
plan
ahead
for
the
resources
that
we
need.
E
A
Thanks
everyone,
and
especially
thanks
clayton.
I
think
that
that
is
all
the
questions
that
we've
had
so
far.
So
just
following
up
on
the
question
about
file
system
licenses
on
perlmutter,
they
are
the
same,
the
same
usages
on
corey
and
the
same
names
as
on
corey.
We
do
need
to
update
the
documentation
so
to
make
that
more
obvious,
but
there's
a
an
extra
sort
of
nicety.
The
pearl
matter
has
which
is,
if
you're
submitting
from
scratch.
A
It
will
make
the
assumption
that
you
probably
want
a
scratch
file
system
license,
and
so
it'll
add
that
in
automatically
so,
for
you
know,
for
a
lot
of
cases,
you'll
sort
of
get
the
get
the
licenses
for
free
if
you're
submitting
from
home.
Knowing
you
need
scratch
you're
going
to
want
to
add
that
in
explicitly.
C
C
C
A
C
Oh
okay,
yeah.
I
tend
to
have
submission
script
in
somewhere
in
cfs,
but
the
actual
data
in
is
scratch
and
input,
data
or
input
text
file
and
sometimes
model
executable,
is
under
a
common
program
or
software
place
or
scotch
space.
So
it's
kind
of
split
around,
but
yeah
it's
dutch
command
is
I
tend
to
issue
from
cfs
right.