►
From YouTube: NUG Monthly Meeting, September 16, 2021
Description
Recording of NUG monthly meeting for September 2021 -- see https://www.nersc.gov/users/NUG/teleconferences/nug-meeting-16-sept-21/ for slides and uploads.
A
Oh
perfect
recording
started
but
a
little
late
I'll.
Take
it
all
right.
The
recording
theory
showed
up
yeah
perfect
all
right.
So,
let's
start
out
with
the
win
of
the
month.
So
has
anybody
have
any
accomplishment
any
achievement?
Have
you
had
a
paper
accepted?
Have
you
submitted
a
solved
bug
on
the
systems?
Anything
like
that
that
you'd
like
to
discuss
bring
up
show
off
now
is
the
perfect
time.
B
Willem
go
ahead.
Just
gave
a
talk
yesterday
on
openacc
summit
conference
about
work
that
we
performed
on
curry,
gpu
me
and
lenormalin.
A
B
It
is
recorded,
so
I
can
give
it
once
they're
available
perfect.
A
D
C
E
Richard
yeah,
so
I
just
put
it
into
chats
that
we
we
just
announced
we're
starting
a
seminar
series
with
speakers
from
the
nurse
early
career
achievement
award
recipients,
and
so
I
just
put
the
url
in
there.
It
looked
like
some
really
interesting
talks,
so
everybody
is
invited
just
wanted
to
bring
that
to
everybody's
attention.
E
Oh
yes,
I
forgot
to
tell
you
kevin,
so
we
can
add
that
to
the
slides.
A
A
Else
all
right,
I
don't
particularly
have
a
particularly
strong
win
this
month
other
than
some
early
pearl
mudder
discussion
points.
So
nothing
big
for
me
to
talk
about
just
yet
here,
it's
nice
to
see
a
lot
of
nurse-related
activities
going
on
out
there
very
good
all
right.
So,
let's
move
on
to
the
next
topic
here.
A
What
you've
possibly
learned
this
month
is
there
anything
that
you've
been
doing
that
was
unique
this
month?
There's
something
new
something
say
a
feature
on
nurse
that
you've
learned
about
or
something
that's
made
your
life
a
little
bit
easier,
one
of
the
more
common
ones
that
pops
up
from
time
to
time
is
the
home
backups.
There
are
actually
backups
for
your
home
directory
at
nursk
and
it
usually
ends
up
saving
somebody,
and
it's
still
remarkable
to
me
that
some
people
don't
know
about
those.
A
A
Let's
see
so,
there
was
an
interesting
post
in
nurse
about
a.
If
you
can't
do
it
on
wait.
You
can
do
it
on
nurse.
You
can.
What
is
it?
You
can
pipe
output
straight
to
your
clipboard,
so
I've
been
doing
that
on
my
local
machine,
a
little
bit
here
to
pipe
things
to
the
clipboard
and
then
mousing
somewhere
else
in
order
to
paste
it
say
into
a
document
away
from
my
terminal,
which
has
been
incredibly
interesting
and
helpful.
A
I've
been
working
on
graph
based
stuff
in
python,
so
I
have
jupyter
hub
open
in
one
box
and
then
the
output
from
the
application
in
the
other.
I
pipe
the
information
to
my
my
clipboard
and
then
just
copy
paste
it
into
the
jupiter
hub
notebook
in
a
tab
somewhere
else.
It's
incredibly
easy
and
fast
I've
liked
it
a
lot.
A
Just
saw
oh
yes,
the
pi
toolbox.
Yes,
if
you
guys
haven't
taken
a
look
at
that,
do
we
have
like
documentation
for
that
laying
around?
We
could
point
them
to
you
very
quickly.
E
I
think
it's
in
my
nerve
skin,
it's
I
think
it's
pretty
intuitive.
We
could
maybe
paste
the
url
in
there.
If
somebody
has
it
yeah,
something
like
that.
A
E
One
of
the
problems
it
was
built
to
address
was,
if
you
have
a
project
directory,
and
you
have
a
big
project
that
people
are
coming
and
going.
People
will
set
the
ownership
of
files
to
something
that
maybe
is
not
ideal
and
then
they're
gone,
and
then
you
need
to
kind
of
reclaim
it
as
the
pi
or
whatever,
and
you
couldn't
do
that
before.
So
this
gives
you
a
way
to
go
in
there
and
change
all
those
those
file
permissions
without
having
to
file
a
ticket
and
get
assisted
men
to
go,
do
something
as
rude.
A
Yeah
through
the
ticket
that
can
be
incredibly
complicated
as
we
have
to
confirm
who
you
are
and
that
you
have
permission
to
change
it.
If
you
have
two
or
three
different
people,
you
want
to
jumble
it
around.
We
have
to
make
sure
it's
all
set
up
right
before
we
do
one
operation,
so
that
is
a
much
quicker
sleeker
way
to
do
that
kind
of
operation.
F
Oh,
can
I
just
ask
kind
of
a
question
this
is
koichi
from
here
now,
so
I
just
learned
actually
finally
how
to
look
at
my
past
jobs,
and
you
know
how
much
compute
hours
and
scholars
I
used
for
from
iris.
F
I've
been
kind
of
doing
for
my
this
from
my
mask
and
is
you
know,
a
kind
of
combination,
so
I'm
kind
of
think
what
would
be
the
role
for
minors
and
iris
like
what
is
your
use
case
for
each
of
those
tools?
Are
they
do
they
have
some
overlap,
or
are
you
kind
of
trying
to
shift
from
one
to
the
other
in
the
near
future?
So.
E
I
can
take
that
one
too
kevin,
so
that's
a
great
question
and
iris
is
our
our
new
kind
of
accounting
system.
It's
really
iris
is
really
the
point
of
truth
for
everything
at
the
center
now
miners
was
developed
in
the
past
to
provide
for
some
very
useful
functionality
that
wasn't
available
anywhere
else,
but
we
realized
that
right
now
that
there's
kind
of
some
things
that
aren't
yet
an
iris
that
are
still
useful
in
in
my
in
my
nerve
skin
and
also
some
places
where
there's
overlap.
E
So
what
we
did
was
we
actually
funded
a
study
from
some
experts
in
user
experience
and
that
sort
of
thing
that
took
a
look
our
whole
ecosystem
and
they
just
gave
us
recently
a
very
nice
report
about
strategies.
We
could
use
things
that
need
attention
and
how
we
could
consolidate
things
to
improve
the
user
experience
and-
and
so
we're
really
going
to
take
that
to
heart.
I
think
and
try
to
work
out
all
these
these
questions,
because
you've
identified
something
that
that
we
recognize
as
well.
A
We
also
regularly
talk
about
updates
to
iris
a
lot.
So
if
you
have
a
suggestion
or
tweak
or
something
you'd
like
to
see
file
a
ticket,
we
discussed
those
very
actively.
I've
discussed
two
or
three
improvements
to
iris
in
the
past
couple
of
months
that
are
hopefully
going
to
get
in
there
sometime
soon
to
make
quality
of
life
a
little
bit
better.
A
All
right,
if
there's
nothing
else
in
what
we've
learned,
we
can
move
on
to
announcements
and
our
presentation
so
announcements
for
the
month.
All
of
these
are
in
the
weekly
email
so
for
details,
just
search
for
rebecca's,
most
recent
weekly
email,
I've
pulled
most
of
the
the
higher
level
ones
out
of
this
week,
but
anything
similar
would
also
be
in
there.
So
just
check
your
weekly
emails
for
all
the
details,
so
ercap
is
now
open,
as
we
will
find
out
today.
A
There
were
a
few
changes
from
previous
years,
so
for
those
of
you
who
filled
this
out,
keep
your
eye
out,
make
sure
that
you
do
everything
appropriately,
so
that
you
we
have
a
very
smooth
transition
process.
We
also
have
office
hours
that
are
going
to
be
available
for
ercap
on
september
30th
and
october
4th.
A
So
if
you
have
any
questions-
or
you
want
someone
to
help
you
through
something-
that's
particularly
confusing
or
convoluted
in
there
or
just
want
to
know
exactly
how
to
answer
a
specific
question
that
best
helps
us
feel
free
to
just
drop
in
and
ask
the
question
we'll
be
certainly
very
glad
to
help
there's
a
better
scientific
software,
2022
fellowship
program
and
the
applications
are
now
open.
They
have
up
to
a
25k
grant
for
those
of
you
who
know
this.
This
is
a
very,
very
solid
program.
A
A
lot
of
the
ecp
applications
are
involved
in
this
in
various
forms
specifically
meant
too
foster
practices
processes,
tools,
things
like
ci
things
like
good
practices
when
you're
using
github
things
like
tools
to
help.
You
make
sure
that
your
code
stays
consistent
or
works
across
an
incredibly
large
number
of
architectures
or
scales.
How
to
build
these
tools,
how
to
make
them
efficient
and
effective.
All
that
kind
of
improved
scientific
software
productivity
stuff
is
applicable
to
this,
so
the
applications
are
open.
A
If
you
have
any
ideas
or
you
have
any
interest
or
anything
along
those
lines,
please
feel
free
to
look
up
what's
required
there.
In
the
in
the
new
user
in
the
weekly
email,
the
women
in
hpc
workshop
is
also
up
and
up
and
coming
I've
been
to
one
of
these.
They
are
extremely
useful.
Please
consider
going
to
this.
The
deadline
is
currently
extended
to
the
end
of
the
month.
A
So,
if
you
have
anything
to
put
in
here
along
the
lines
of,
we
have
the
description
down
there
of
the
of
the
types
of
things
that
are
included.
Anything
early
in
mid-career
allies
are
allowed
to
give
very
interesting
talks
about
things
that
they're
doing
in
there
things
that
they've
seen
or
things
they'd
like
to
help
with
or
even
how
allies
can
best
contribute
people
responsible
for
hiring
and
recruiting
to
make
sure
that
there's
no
bias
in
that
there's
a
diverse
and
fair
environment.
All
that
kind
of
stuff
is
considered.
A
It
is
incredibly
informative.
Please
consider
either
going
or
or
submitting
something
to
the
to
the
workshop
that
you
have
two
weeks
left
to
contribute
to
that.
So,
if
you're
interested,
please
please
do
that
and,
as
you
might
also
imagine,
with
perlmutter
there's
an
absolute
gigantic
number
of
trainings
and
types
of
things
going
on.
So
the
typical
july
gpu
hackathon
this
year
got
so
many
teams
applying
to
it
that
we've
decided
to
do
another
one.
So
there's
now
a
december
gpu
hackathon
and
the
applications
are
open.
A
They're
due
october
8th
that
you
can
find
the
details
and
the
links
in
in
the
weekly
email,
the
only
you
might
see.
The
name
is
a
bit
weird
here,
so
the
hackathons
are
technically
ran
by
the
open
acc
organization,
which
is
a
non-profit
organization.
That
nvidia
helps
out
a
lot,
but
it
isn't
nvidia.
So
nvidia
sends
a
lot
of
mentors
to
these
events,
but
technically
speaking,
it's
the
open,
acc
organization.
That
runs
it.
So
it
ends
up
a
bit
confusing,
but
it's
really
and
nvidia's
on
board
heavily
to
with
these
events.
A
So
if
you're
reporting
or
you
want
support
or
you're
interested
in
seeing
what
will
be
required
or
you
have
some
interesting
nasty
performance
things,
you
need
help
with
feel
free
to
apply.
I'm
actually
the
person
running
that.
So,
if
you
have
any
questions
about,
say
the
application
or
the
email
or
what
you
need
to
do
to
have
a
really
strong
application
feel
free
to
just
email.
Me
I
can.
I
can
answer
all
your
questions
with
that,
and
these
are
just
a
few
of
the
trainings
that
are
coming
up.
A
We
have
an
openmp
device
offload,
an
e4s,
deep
dive
and
an
ai
for
science
bootcamp
in
the
next
couple
of
weeks.
There's
cuda
training
series
has
multiple
additional
things.
If
you
saw
that
last
year,
that
are
also
coming
up,
there's
a
very,
very,
very,
very
long
and
detailed
list
of
good
trainings
that
should
be
coming
up.
A
So,
if
they're,
if
anybody's
interested,
feel
free
to
find
the
particular
link
and
register,
should
be
pretty
all
these
are
very
interesting
and
which
ones
I
can
go
to
really
comes
down
to
schedules
and
meetings
and
what's
critical,
so
I'll,
hopefully
I'll
see
a
lot
of
people
at
a
lot
of
them.
Does
anybody
else
have
any
other
announcement
for
anything
coming
up
that
they
would
like
to
share
with
the
group.
A
Thank
you
yep.
So
if
you
want
to
apply
the
hackathon
it's
there,
the
link
is
also
in
the
weekly
email.
So
you
can
also
just
go
to
the
weekly
email
and
find
the
item
for
that,
and
that
should
also
have
the
links
there
as
well.
A
All
right!
Oh
for
that
one.
There
was
also
those
just
a
question
on
perlmutter.
As
long
as
perlmutter
isn't
down
for
maintenance
or
something
you
will,
we
will
have
access
to
pearl
mudder
for
that
event,
links
for
the
ai
for
science
bootcamp.
They
should
also
be
in
the
weekly
email
willem.
Do
you
think
you
can
grab
that
one.
A
A
If
I
had
two
monitors
I'd
do
it
myself,
but
in
my
current
setup
I
have
exactly
one
so
you
can
see
a
screen
or
I
can
put
things
in
chat.
Oh
you
only
get
one
all
right,
okay,
so
with
that,
I
believe
we
have
the
general
introductions
and
announcements.
I
hope
everyone
is
making
good
research
progress
at
nursk.
As
always,
if
you
have
any
questions,
if
you
need
any
help,
feel
free
to
send
us
a
ticket
feel
free
to
look
through
the
documentation,
all
the
documentation.
For
those
who
don't
know
now
takes
user
input.
A
We
now
have
a
forward-facing
github,
so
if
you
see
something
in
the
documentation
that
is
wrong
or
you
would
like
to
add
something
somewhere
that
you
think
would
be
useful
or
you
want
us
you've
seen
something
that
you
have
a
document
for
laying
around
and
you
just
like
to
have
a
permanent.
This
is
where
this
is
feel
free
to
apply.
Thank
you
very
much
willem
perfect,
all
right.
Okay,
anything
else
before
we
go
on
to
our
presentation
by
clayton
anything
to
announce
or
talk
about.
A
All
right
sounds
like
it's
time
to
talk
about
ercap.
Are
you
ready,
clayton.
A
And
so
for
everybody
who
does
not
do
our
cap,
do
you
have
a
brief
introduction?
What
ercap
is
I've
generally
found
that
information
like
that,
even
though
I
don't
do
it
is
extremely
useful.
You
get
an
idea
for
what's
happening,
what's
important,
what's
critical,
where
people
are
moving
and
going
with
things,
so
I
think
it's
seeing
what
the
eurocap
is
be
very
useful
for
a
large
group
of
people
and
understanding
what
ercap
is
generally
might
be
good
for
everybody
here
who
doesn't
fill
it
out?
G
G
So
this
is
this
is
where
researchers
apply
for
resources
here
at
nurse.
That's
the
computing
time,
the
storage,
time
etc.
G
G
G
So
for
this
year
we
actually
have
a
little
bit
of
a
shortened
ercap
season.
If
you
want
to
call
it
that
we
opened
up
ercap
on
september
7th
right
after
labor
day
and
all
our
cap.
Submissions
for
2022
are
due
by
october
4th.
G
We
do
collect
aircap
requests
year-round,
but
the
vast
majority
of
them
are
submitted
at
this
time.
And
if,
if
you
don't
get
your
aircap
request
in
by
october
4th
and
it
gets
evaluated
by
the
doe
allocation
managers,
you
may
get
left
out
of
being
able
to
have
time
for
2022.
That's
why
you
need
to
be
able
to
try
and
get
everything
in
by
october
4th.
G
G
So
for
2022
we're
changing
our
unit
of
currency.
If
you
want
to
call
it
that
currently
we
we
charge
in
what
we
call
nurse
hours,
but
next
year
we're
going
to
be
changing
that
to
what
we're
going
to
be
calling
node
hours.
G
Cpu
node
hour
is
essentially
running
for
one
hour
on
one
cpu,
only
node,
now
the
node
hour,
the
cpu
node
hours
can
be
used
on
both
promoter
and
on
corey
and
corey
has
scaling
factors
for
if
you
run
on
knl
or
haswell,
and
then
for
the
gpu
node
hour.
Again,
that's
running
for
on
a
single
promoter,
gpu
accelerated
node
for
one
hour.
G
Some
conversions
factors,
kind
of
so
you
know
we
run
a
lot
of
performance
tests
and
there's
some
scaling
for
the
corey
cpu
nodes.
So
like
a
haswell
node
is
a
0.34.
G
Performance
factor
versus
a
pro
motor,
cpu
and
and
knl
is
0.2,
but
when
figuring
out
how
many
node
hours
to
request
in
the
ercap
request,
if
you
look
at
your
current
usage
and
essentially
kind
of
divide
it
by
400,
that's
what
you
would
be
requesting
for
for
next
year.
So
in
our
example
here
say:
if
you
used
a
million
nurse
hours,
you
divide
that
by
400
and
then
you
would
request
2500
cpu
note
hours
for
next
year.
G
There
any
questions
on
that.
I
know
it's
kind
of
a
little
rough,
but
I
don't
I
I
don't
see
the.
I
don't
see
the
chat,
so
I
don't
know
if
anyone's
asking
any
questions
or
not.
G
Okay,
so
for
next
year,
the
available
cpu
time
will
be
about
24.6
million
cpu
note
hours
and,
of
course,
those
are
distributed
amongst
the
various
ue
office
of
science
programs,
and
I'm
not
sure
what
this
link
is
here.
Richard
did
you
add
that.
E
Yeah,
that's
the
this
actually
goes
to
the
call
for
the
call
for
proposals,
but
it
has
the
table
on
there
that
says
how
much
time
is
available
to
allocation
managers
in
total
for
your
area.
So
it'll
give
you
some
sense
to
let
you
know
that
you're
you're
not
requesting
more
than
they
have
or
100
of
what
they
have
or
just
to
kind
of
help.
You
get
some
order.
Magnitude
feel
I
don't
think
we
say
in
this
presentation,
but
as
far
as
the
cpu
node
hours
for
2022.
E
they're
about
if
I
recall,
30
to
40
percent,
more
availability
than
there
is
for
this
year,
so
that
can
also
help
you
size.
Any
cpu
requests
that
you
might
want
to
make.
E
So
we
don't
know
yet
when
charging
will
begin
on
perimeter,
so
we're
we're
being
rather
conservative.
I
think,
and
if
things
go
better
than
expected,
we
might
have
to
have
some
kind
of
additional
allocation
augmentation
for
next
year.
I
mean
that
would
be
great
for
everybody.
I'm
sure.
G
A
E
Oh,
I
don't
think
jack
is
on,
but
our
performance
team.
You
know
we
have
some.
We
have
some
of
the
same
cpus
that
will
be
in
phase
two,
this
system,
and
so
we
run
it.
We
ran
a
bunch
of
popular
apps,
a
bunch
of
some
of
the
nisap
apps
cpu
only
and
did
a
comparison,
the
performance
versus
haswell
and
knl,
and
this
is
kind
of
a
rough
average,
and
so
it
might
be
that
one
system
or
the
other
performs
better
for
your
code
with
this
scaling
factor.
E
But
because
the
hours
are
interchangeable,
you
can
run
either
on
haswell
or
k,
l
and
corey,
or
on
the
promoter
cpu
only
nodes
to
whatever
optimizes,
whatever
you're,
trying
to
optimize
the
best,
whether
that's
cost
or
throughput,
or
performance
or
latency,
to
start
a
job
whatever
works
out.
So
it
gives
you
some
freedom
to
be
able
to
try
to
to
move
amongst
the
systems
and
find
what's
best
for
you,
but
that's
a
that's
kind
of
a
that's
kind
of
a
global
average
across
a
little
conservative.
I
think
global
average
across
the
apps
that
we
tested.
A
I
would
agree,
I
think
I
saw
something
a
little
higher
than
that
when
we
ran
so
I
think
that's
pretty
fair
metric
one
other
question
that
I
actually
have
so
on
the
on
that
linked
page,
the
2020
call
for
proposals
to
use
nurse
resources.
I
put
that
in
the
chat
the
the
mini
link.
I
just
expanded
it
put
that
in
the
chat
for
anyone
who's
interested.
It
says
under
the
table
that
you
should
put
in
hours
for
perlmutter
assuming
the
entire
year.
Is
that
correct?
A
That
seems
to
run
a
little
contrary
to.
We
might
need
to
augment
things
down
the
road.
E
I
think
you
should
you
should
request
what
you
think
you
will.
You
will
need
for
the
year
and
then
the
program
managers
have
a
certain
amount
and
so
they'll
they
will
scale
it,
but
at
least,
if
you
put
in
the
full
year
amount,
if
we
do
do
an
augmentation,
then
that
could
be
a
consideration
on
how
that
argument.
Augmentation
is
distributed.
G
Okay,
so
also
for
gpu,
we
have
a
3.4
million
gpu
note
hours
that
we're
planning
on
distributing
for
next
year
and
there's
also,
let's
see
what
is
this
applications?
G
E
So
some
of
you
run
community,
or
you
know,
common
hpc
applications
like
espresso
vasp,
particularly
in
in
chemistry
and
materials,
and
condensed
matter
physics,
and
so
we're
trying
to
compile
we're
in
the
process
of
compiling
a
list
of
the
most
popular
ones
of
those
that
we
know
have
a
gpu
version
and
we're
putting
at
this
page.
This
page
is
still
kind
of
rough,
but
over
the
next
week
I
think
it
will
be
more
informative.
C
I
have
a
question:
if
you're
building
your
own
application
and
you've
worked
hard
to
get
it
running
on
gpu.
How
much
evidence
do
we
need
to
put
in
your
cap
that
we
can
run
on
the
gpus
and
also
is
I
thought
the
aircap
is
decided
by
the
program
managers?
Who
may
not
be
the
expertise?
C
E
Right
so
we're
we're
asking
you
all
to
be
persuasive
to
the
program
managers.
So
if
you're
part
of
nisap,
for
instance,.
E
Making
that
statement
is
certainly
a
good
one
if
you've
run
on
another
gpu
system
somewhere,
and
you
have
evidence
that
it
works.
Of
course
that's
very
strong
evidence
to
that
effect.
If
you
filled
out
kevin's
our
form,
I
say
it's
kevin's
form
because
I
haven't
put
it
up.
Kevin
has
a
form
which
I'm
sure
would
be
happy
to
point
you
to
that
asks
you
some
questions
about
your
gpu
readiness
and
we
have
that
information.
We
can
share
with
them
as
well.
C
G
And
also
when
we,
when
we
get
to
the
actual
cap
form,
when
you
get
to
the
section
for
resources
and
you're
requesting
gpu
time,
you
will
need
to
include
a
statement
of
your
code's
readiness
to
use
gpu.
And
you
can
put
your
explanation
there,
that
you
run
it
on
other
gpu
systems,
and
you
know
give
some
backup
information
that
that
it
is
ready
to
to
run.
E
Okay,
so
so
josh
has
a
very,
very
good
point
in
chat.
If
I
shouldn't
speak
out
tournament
kevin,
maybe
you're
the
best
representative
here
we
do
have
a
small
a100
system
attached
to
corey
is
that
could
that
be
made
available
to
people
kevin?
I
forget
how
we're
running
that
right
now
for
the
moment
actually,
dgx's.
A
Currently
there
isn't
anything,
stopping
it.
Moaz
is
the
one
currently
in
charge
of
it.
I
think
the
biggest
problem
there
is
it's
exactly
two
nodes,
so
it's
very,
very
small
and
tiny.
So
you,
you
know
the
who
gets
access.
If
everyone
had
access,
no
one
be
able
to
do
anything
on
it.
So
we
have
to
be
a
bit
careful
about
who's,
who
we
put
on
there
and
for
how
long
people
go
on
there.
A
E
Know
one
thing
we
can
do
kevin
I
haven't
thought
of,
and
so
I
would
make
a
note
of
this
is,
I
think
we
could.
We
could
have
the
performance
team
share.
Perhaps
some
average
experiences
on
how
much
faster
things
are
running
on
the
a100s
versus
the
b100s
and
we
do
have
a
larger,
not
much
larger,
but
we
have
a
larger
v100
partition
that
I
think
anybody
can
get
access
to.
E
So
perhaps
we
could
advertise
that
a
bit
more.
Let
me
let
me
talk.
Let
me
talk
to
jack
and
everybody
and
see
if
we
can
make
some.
A
Statement
about
that,
that
might
be
a
pretty
fair.
Like
a
you
know,
final
use
for
corey
gpu
and
the
dgx
boxes
is
you
know,
initial
performance
measurements
for
this
type
of
stuff.
I
believe
that's
quite
fair
for
v100
to
a100.
I
can
give
a
general
rough
number
now
for
most
applications
that
are
doing
are
should
experience
pretty
smooth
transition.
The
numbers
they're
seeing
are
somewhere
between
a
1.5
to
a
2x
increase
from
b
100,
a
100.
very,
very,
roughly,
very,
very
wrong.
There
you
go
so,
but
that's
a
pretty
good
place
to
make.
G
G
G
And
now
just
kind
of
a
rough
quick
run
through
what
the
form
itself
looks
like
and
what's
what
we're
requesting.
So,
when
you
log
into
ercap.nurse.gov
it
should
take
you
to
this
ercap
request.
Homepage
there's
a
couple
of
so
we
have
a
on
the
left
hand,
side.
There's
the
blue
navigation
bar.
G
If
you
don't
end
up
at
this
page
when
you
log
in
say,
because
you
can
get
there
here
also
by
just
logging
into
help.course.gov,
there's
the
manage
my
requests
link
on
the
left
hand.
Side
will
take
you
to
this
page.
It's
divided
into
four
sections.
So
at
the
top
there's
the
buttons
to
start
new
requests
or
renewal
requests
and
once
you've
started
a
request,
it
will
end
up
in
the
draft
request
section.
So
you
don't
have
to
finish
it
all
at
one
sitting.
G
G
It
down
in
the
request
under
review
section,
and,
what's
maybe
important
for
most
of
you-
is
the
previous
requests.
You'll
see
all
of
your
previous
requests
that
have
been
completed
and
provisioned
and
if
you're
going
to
do
a
renewal,
you'll
need
to
look
at
the
ercap
number.
From
your
last
request
to
to
add
to
to
link
to
your
renewal
request,
renewal
requests
will
copy
over
information
from
your
previous
request
and
fill
out.
Some
of
the
information
for
you.
G
G
G
Once
we
get
into
the
forum,
we
do
try
to
help
you
out
with
a
lot
of
different
hints
and
and
things
and
helpful
texts
in
different
colors
and
stuff.
G
As
you
see
here,
the
the
pink
one
says,
if
you're
doing
a
renewal
request
you
have
to
put
in
the
the
or
cap
number
of
the
requests
that
you're
going
to
renew.
That's
why
you
need
to
find
that
in
the
previous
page.
G
G
G
G
You
can
also
look
search
by
keywords,
so
if
you
put
in
science
or
physics
or
something
like
that,
it'll
shorten
automatically
shorten
the
list
for
you
and
if
you
want
to
get
the
full
list,
there's
a
these
fields
have
a
little
magnifying
glass
next
to
them.
On
the
right-hand
side.
If
you
click
on
that
magnifying
glass
it'll,
give
you
a
pop-up
form
that
you
can
go
through
and
do
additional
searches
and
and
look
for
things.
G
G
Okay,
all
right
now
going
back
to
the
the
different
tabs.
We
have
a
personnel
tab,
so
it
has
basically
two
sections
senior
investigators.
Those
are
key
people
that
will
be
on
your
project
that
you
want
to
give
them
recognition
and
such,
but
we
also
have
authorized
preparers.
So
if
you
have
a
big
project
and
a
number
of
people
are
collecting
information
for
your
request,
you
can
have
add
them
as
authorized
preparers
and
they
can
go
in
and
help
you
edit
the
request.
G
G
Okay,
I.
G
C
I
thought
in
the
past,
pi
proxies
were
automatically
also
authorized
to
help
preparing
the
error
cap.
G
When
we
were
using
the
old
nem
system
and
and
aircap
was
tightly
integrated
with
them,
we
could
do
that,
but
when
we
went
to
a
new
system
and
brought
in
iris
we
we
don't
have
that
that
tight
integration
between
user
roles
and
projects
and
such
so
pi
proxies
are
not
automatically
created
as
authorized
preparers
and
vice
on
the
flip
side.
If
you're
an
authorized
preparer,
you
don't
automatically
become
a
pi
proxy
when
the
project
is
provisioned
over
an
iris
either.
G
Right
if
you're,
not
already
on,
like
the
the
current
or
cap
request
as
an
authorization.
G
G
G
D
G
Okay
funding
so,
of
course,
the
major
funding
sources,
the
ue
office
of
science,
and
when
you
click
on
this
check,
box,
you'll
need
to
select
the
the
program.
That's
providing
the
funding
the
program
manager
that
was
involved
in
your
grant
and
you
can
select
from
a
list
of
the
the
designated
program
managers
that
we
know
about,
and
then
there's
also
a
field
for
including
the
of
science
grant
and
or
pam's
or
fwp
numbers.
Whatever
is
appropriate.
E
So
let
me
introduce
something
right
here
is
that
you
know
we
we
talk
with
the
program
managers.
E
A
lot,
and
one
thing
that
has
become
abundantly
clear
to
me
is
that
they
take
this
section
extremely
seriously
and
if
you
put
in,
if
you
fail
to
put
in
a
grant
number,
if
you
say
you're
the
office
of
science
fund
and
you
fail
to
put
in
a
grant
number
or
you
put
in
a
wrong
grant
number
or
a
wrong
funding
program
manager,
in
other
words
the
program
manager
that
is
on
your
research,
grant
it
it
it's
it's
not
it's
not
as
good
to
those
players,
it's
not
as
good
for
the
proposal
as
if
you,
if
that
stuff
was
filled
in
accurately.
E
A
We
just
had
a
comment
that
it
might
be
up
richards
alley
as
well,
so
they
have
do
you
see
the
chat?
It's
probably
easier
than
reading
this
one,
this
one's
a
bit
technical.
E
Oh
hi
hi
welcome
welcome
to
our
chat
I,
if
you're
still
in
india,
I'm
not
sure
what
time
it
is.
Hopefully,
it's
not
the
middle
of
the
morning.
E
So
this
is
the
the
the
the
covid
19
research
special
program
is
actually
funded
outside
of
pretty
much
the
everything
we've
been
talking
about
today.
It's
part
of
the
the
nurse
director's
reserve
program,
and
so
it
could.
It
should
continue
to
be
a
nurse
director's
reserve.
E
Yeah,
the
consortium
is
not
exceed
per
se,
so
I
admit
it's
pretty
confusing,
because
the
consortium
does
use
a
lot
of
the
infrastructure
and
tools
developed
by
xc.
So
yeah
just
continue
to
put
it
as
a
director's
reserve
program
in
the
what
we
call
the
the
director's
reserve
campaign
category
or
program.
B
E
G
And
as
far
as
the
user
interface
functionality
you'll
see,
there's
the
red
asterisks
there
next
to
funding,
program,
manager
and
grants
and
stuff
now
the
red
text
up
there
that
says
you
must
select
at
least
one
funding
source.
That's
going
to
stay
red,
no
matter
what
you
do.
So
that's
that,
unfortunately,
we
don't
have
a
way
to
change
the
color
on
that
text
when
all
of
the
information
is
requested,
but
some
people
have
been
thrown
off
by
that
and
say
well,
I've
filled
in
everything,
but
it
still
says
this.
G
I
have
to
select
this
one
funding
source
yeah,
that's
just
kind
of
the
way
this
system
works
and
you
can
ignore
it
or
that
until
you
know,
as
long
as
you've
selected
everything
that
you
need
now,
besides
the
did
it
change
there.
It
is.
G
Okay,
besides
the
oe
office
of
science,
we
also,
of
course,
many
programs
are
funded
by
other
agencies,
so
other
federal
agencies
other
than
the
doe
office
of
science.
You
can
include
their
information
under
under
here,
and
you
can
in
this
section
you
can
actually
select
multiple
agencies
to
include
so
like
here.
We've
got
the
national
science
foundation
and
nih
or
human
health
and
human
resources.
G
G
If
you
have
some
funding
like
this,
that
hasn't
gone
through
doe
already,
you
do
have
to
include
some
type
of
explanation
of
how
your
project
supports
the
mission
of
the
doa
office
of
science
that
you're
sending
your
proposal
to.
E
G
Yeah,
so
once
you,
if
you
select
view
office
of
science
up
above,
this,
will
actually
be
grayed
out,
so
you
don't
have
to
put
anything
there.
G
And
did
it
change
here?
We
go
acting
kind
of
slow
today.
So,
as
far
as
security
goes,
nurse
only
supports
open
research.
G
Basically,
it's
a
yes
or
no
question
as
long
as
you
don't
need
to
use
any
of
this
stuff.
You
can
just
say
yes,
however,
if
there's
something
that
you're
doing
that,
you
need
to
request
an
exception,
for
you
can
select
the
the
no
and
give
us
a
reason.
Why
and
that'll
be
taken
into
consideration
about
whether
we
can
support
that
or
not.
G
G
G
There
we
go
project
details,
so
we
actually
asked
for
two
different
explanations
of
the
project,
so
the
first
one
is
kind
of
the
generalized
ameri
scientific
american
level,
description
of
what
your
project
does
and
what
you
hope
to
accomplish
and
then
there's
a
more
detailed
description
that
you
want
to
give
to
doe
for
what
your
aim
is,
the
more
technical
stuff
of
how
you
plan
to
do
your
research
and
what's
required,
etc.
E
Yeah
so
clayton,
I'm
gonna
make
a
real,
quick
comment
on
this
too,
so
I
we
consider
these
both
important,
also
so
for
the
first
one.
This
is
the
first
place
we
go
when
we
see.
Oh,
we
we
notice,
oh
there's
a
really
cool
paper
in
jacks
or
whatever
that
that
that
acknowledged
nurse.
This
might
make
a
good
science
story
or
something.
G
Okay
and
then
also
under
this
tab
for
renewals,
we're
going
to
ask
you
for
your
summary
of
your
accomplishments
for
the
past
year
of
using
nurse
and
how
we've
helped
you
do
what
you
do
and
we
are
still
collecting
your
publication
information
here.
G
E
Yeah
so
I'll
make
another
comment
here
too
clayton,
so
we
have
developed
a
way
to
track
publications
in
iris
and
we'd
hoped
it
would
be
ready.
So
we
wouldn't
have
to
have
you
fill
out
this
box
this
year,
but
it's
not
quite
so
we're
going
to
ask
you
to
do
it
again
next
year.
We
hope
that
we
will
be
all
in
iris.
E
E
If
we
don't
have
a
strict
rule
around
this,
but
if,
if
you
acknowledge
us
in
the
acknowledgments
by
for
us,
is
put
the
word
nurse
or
national
or
spell
us
out
completely
we'll
find
it
and
those
certainly
get,
I
should
say
more
weight.
I
guess
in
the
sense
that
we
we
can
look
up,
we
can
look,
we
can
look
up
from
crossref
or
whatever
you
know
what
the
paper
was
when
it
was
published,
and
all
this
sort
of
thing.
E
E
So
there's
not
a
there's,
not
really
a
great
answer
about
this,
but
you
know,
through
through
nuggets
and
through
nug,
we've
been
having
some
of
these
discussions
too
about
what
is
the
best
way
to
markup
or
tag
or
whatever
your
your
publications,
so
that
we
can
find
them
and
that
they
they
have
some.
I
guess
authenticity
as
it.
C
E
E
If
you
put
an
accurate
doi
in
here,
we
will
be
able
to
automatically
validate
it
against
dois
that
that
we
know
acknowledge
nurse
in
the
text
of
their
paper.
Okay,
if
you
don't,
then
we
may
or
may
not
count
for
various
things,
although
we
will
show
it
to
the
program
managers,
but
they
might
get
frustrated
if
they
can't
easily
validate
it
too.
E
G
Okay
and
now
probably
the
most
important
section
for
you
folks,
is
the
resources
requested
tab.
So
here
we're
showing
you
what
the
cpu
node
hours
that
you've
used
this
year
and
then
you're
going
to
request
the
number
cpu
node
hours
for
next
year.
Now
these
fields
for
requests
for
the
cpu
and
gpu
node
hours.
They
are
integers,
so
you
can't
type
in.
I
want
2
million
hours
for
next
year
or
even
2000
or
200.
G
Some
a
few
people
have
been
kind
of
stung
by
that
in
the
past
year
and
they
didn't
notice
that
when
they,
after
they
put
in
the
number
that
it
truncated
on
them.
Also,
if
you
use
a
decimal
number
say,
I
only
need
two
and
a
half
or
let's
say
10
and
a
half
note
hours
it's
going
to
truncate
it
down
to
just
either
11
or
10,
depending
upon
your
decimal
point
placement,
etc.
G
We
have
a
couple
of
well.
The
the
readiness
word
there
in
blue
is
a
link.
That'll.
Take
you
to
this
page.
It
talks
about
our
our
readiness
guidelines
that
we
have
there
so
that'll
give
you
some
some
information
that
will
help
you
try
and
determine
what
to
include
in
this
section.
E
Okay,
let
me
comment
on
this
too.
So
we're
going
to
give
everybody.
If
you
only
request
gpu
note
hours
or
if
you
only
request
cpu
hours,
we're
going
to
give
everybody
a
minimum
on
both
just
so
they
can
log
in
or
they
can
explore
one
or
the
other
in
case
they're,
maybe
not
gpu
ready.
They
can
try
some
things
so
that
minimum
is
going
to
be
in
a
100
on
each.
E
So
if
you're
requesting
100
note
hours
or
less
of
either
you
can
just
say
you
don't
need
to
make
a
statement
of
gpu
readiness.
E
I
mean
you
could
just
you
could
just
say
this
is
we
need
to
put
we
need
to.
We
need
to
publicize
this
more
clayton,
so
we
need
to
put
this
on
documentation.
You
can
just
say
this
requests,
100
hours
or
less
okay
yeah.
We
need.
We
need
to
document
that,
but
that,
but
that's
what
we're
going
to
do
is
we're
going
to
give
everybody
at
least
100
on
both.
G
Okay,
of
course,
the
other
information
we're
going
to
need
to
request
is
for
the
hpss
archival
storage
and
the
cfs
project
storage
space.
G
Let's
see
and
still
on
this
tab,
there's
an
overall
justification
for
your
request.
You
want
to
give
an
explanation
of
how
you
came
up
with
determining
how
many
hours
you
needed
how
much
storage
space
you
needed.
G
Okay,
we're
asking
for
your
top
five
codes
that
you
use,
and
basically
the
name
url
of
the
code
website
and
a
brief
description
and
a
new
item
this
year
is:
let
us
know
if
this
code
is
gpu,
enabled,
if
you're
able
to
use
it
on
gpus
or
not.
G
You
can
also
add
attachments
to
your
request
if
you
have
a
longer
description
of
either
your
your
project,
description
or
your
accomplishments
or
your
publications,
anything
additional
information
that
doesn't
fit
in
the
boxes
that
we
have
now,
you
can
add
them
here
and
actually
to
add
the
add,
an
attachment.
It's
it's.
Actually,
a
paper
clip
at
the
top
of
the
page,
you'll
see
a
button
that
says
like
submit
or
save
as
draft
and
then
to
the
left
of
it.
You'll
see
this
little
paper
clip
you
click
on
that.
G
G
The
the
attachments
will
appear
in
two
places
up
at
the
very
top
of
the
page
where
it
says
manage
attachments
and
then,
of
course,
on
the
supporting
information.
The
the
list
of
attachments
is
there
as
well,
once
you've
filled
everything
out
and
you
want
to
kind
of
keep
something
for
reference.
You
can
create
a
pdf
just
by
clicking
on
that
button
and
it'll
automatically
save
the
pdf
under
the
attachment
sections,
and
then
you
can
download
it
to
your
your
computer
to
first
for
safe
keeping.
G
The
last
thing
we
ask
for
is
that
you
do
agree
to
monitor
the
usage
of
your
resources
and
make
sure
that
it
is
for
the
project
that
you
have
requested
this.
So
you
have
to
monitor
what
your
people
are
doing,
make
sure
they're.
Not
you
know
trying
to
mine,
bitcoin
or
something
like
that.
G
And
then
you
submit
your
request:
there's
a
submit
for
review
button.
It's
going
to
ask
you
to
confirm
that
you
want
to
do
this
and
then,
when
you
say
okay,
the
system
is
going
to
go
through
and
look
for
all
the
red
asterisks
that
say
that
you
forgot
to
fill
in
something
and
it'll
give
you
it'll
either
give
you
a
list
of
the
fields
that
are
mandatory,
that
you
left
unfilled
out
or
if
there's
only
one.
It
usually
takes
you
to
that
particular
field
and
then,
once
you've
successfully
submitted
your
request.
A
There
thank
you
very
much,
clayton
sure,
there's
one
here:
where
do
we
find
already
funded
projects.
I
So
I'm
actually
interested
in
the
projects
that
are
already
funded
by
department
of
energy,
so
that
are
already
in
progress
or
have
been
done.
E
E
Dewey
gives
out
research
grants
to
researchers
who
write
proposals
and
then
they
they
study
whatever
it
is
they
want
and
then
and
then
we
have
the
nurse
computational
projects
which
are
separate
things,
I'm
interested
in
research
grounds.
Well,
there's
this
thing
called
pams
that
the
office
of
science
runs.
Let
me
see,
I
think
I
can
put
the
link
in
there
that
in
theory,
you're
supposed
to
be
able
to
look
up
all
these
things.
Let
me
find
that
for
you
in
practice.
E
I
have
trouble
with
it
I'll
put
the
link
in
the
chat
for
here
in
just
a
second,
and
so
you
can
try
to
use
it
and
if
you
have
problems
as
well,
if
you
maybe
just
file
a
trouble
ticket
with
them
or
click
on
pam's
help.
E
H
Anything
else
yeah.
I
still
wanted
to
ask
a
little
more
about
this
factor.
400,
I
presume
that's
based
on
some
earlier
platform
than
corey,
because
you're
saying
corey
is
only
a
factor
of
three
to
five
slower
than
per
meter.
From
your
earlier
view,
graph.
E
Yeah,
that
was
that
was
a
set
of
apps,
so
common
apps
and
some
kneesap
apps
that
were
running
just
on
cpu
that
we
ran
just
on
cpus
that
we
ran
on,
admittedly,
a
single
core
mulan
and
tried
to
make
extrapolation
to
the
dual,
the
dual
socket
phase
two
system.
E
So
it's
it's,
it's,
admittedly,
probably
a
rough
estimate
and
we
were
trying
to
be
a
little
bit
conservative.
So
it's
kind
of
the
best
we
could
do
with
the
current
state
of
things.
E
So
if
you
go
into
iris
and
look
you'll
see
a
number
of
different
units,
one,
let
me
give
you.
Let
me
look
in
here
right
now,
for
instance,
there's
something
we
call
raw
hours,
that's
actually
note
hours,
but
there's
something
we
call.
E
I
should
know
this
off
the
top
of
my
head,
but
I
think
we
call
it
machine
hours
in
iris.
Let
me
do
a
quick
report.
E
Okay,
so
there's
something
called
raw
hours
that
is
actually
note
hours.
On
whatever
note
it
is.
That
has
no
that's
nothing,
but
I
ran
this
many
hours
on
a
node
of
whatever
kind.
It
happened
to
be
machine
hours
has
the
various
performance
factors
factored
in
and
your
charge
would
be
machine
hours
if
you
didn't
run
in
any
special
queue
like
premium
or
low,
or
something
like
that,
so
machine
hours
are
essentially
what
we're
calling
nurse
scholars.
H
Yeah,
I
think
I
understand
that.
Okay,
thanks.
C
E
Yeah,
so
we
haven't,
we
haven't.
Actually
we
haven't
actually
addressed
that
issue,
so
we
will
continue
to
to
have
that
honor
that
that
factor
on
corey
and
what
happens
is
once
we
usually
apply.
Things
like
this
to
we've
always
applied
things
like
this
to
our
largest
production
system,
so
once
perlmutter
becomes
that
at
some
point,
we'll
have
to
make
some
decision,
but
I
would
not.
A
All
right,
so
I
know
we're
over,
so
I
will
blow
through
the
final
few
things
rather
quickly.
Let
everyone
get
to
lunch
so
coming
up
october,
there
is
a
nug
annual
meeting
in
lieu
of
this
monthly
meeting,
so
watch
for
the
date
and
the
agenda
upcoming
november,
there's
tentatively
a
presentation
on
work
from
some
of
our
regular
participants.
A
So
if
there's
any
other
topics,
anyone
would
like
to
see-
or
these
are
of
particular
interests
and
you'd
like
to
hear
them
sooner-
feel
free
to
go
to
the
nurse
user
slack
channel
under
webinars
or
really.
You
can
also
post
it
in
other
places,
and
we
can
point
you
in
the
right
right
direction
to
get
it
in
a
good,
tight
spot,
but
feel
free
to
join
that
nurse
user
slack
and
tell
us
exactly
what
topics
you'd
be
interested
in
coming
all
right,
so
numbers
for
the
system
so
for
the
last
month.
A
So
we
had
a
couple
of
outages
actually
with
a
very
wide
arrange
of
issues,
only
totaling
about
10
hours
of
unscheduled
downtime,
which
is
quite
good
and
an
overall
availability
that
was
sitting
in
the
high
90s
consecutive
consistently
with
pretty
much
exactly
what
was
scheduled
so
overall
overall
last
month
was
pretty
good
and
then
utilization
sat
at
94
percent
with
large
jobs
about
13.
A
That's
a
pretty
good
number
for
large
jobs.
Actually,
tickets
are
coming
in.
D
E
Notes,
1024
notes
and
interesting:
that's
actually
a
very
low
number
and
it's
kind
of
an
outlier
for
the
year
and
and
in
september
we're
back
up.
I
think
over
30
percent
again.
A
Anyway,
maybe
I
read
it
wrong:
maybe
that
has
to
do
with
the
one
I
added
down
here
that
I
found
it
was
quite
interesting
all
right,
so
our
our
ticket
queue
is
sitting
pretty
solid,
so
we're
handling
them
as
they
come
in
pretty
well
slight,
slightly
backlogged.
Of
course
you
get
the
really
complicated
tickets
that
end
up
taking
a
couple
of
months
that
end
up
stacking
up
a
little
bit.
So
that's
about
is
expected
and
clear,
and
this
was
one
of
the
statistics
that
I
saw
that
was
rather
interesting.
A
So
for
last
month
the
amount
of
time
sitting
in
the
queue
on
average
actually
increased
quite
dramatically
last
month.
Maybe
this
is
related
to
this
large
jobs
thing,
but
in
july,
haswell
was
only
a
33
hour
wait
on
average,
whereas
last
month
it
was
192
quite
a
gigantic
leap.
My
current
takeaway
from
this
was,
if
you
could
use
knl,
definitely
use
the
k
l's.
A
You
will
be
through
the
queue
a
lot
faster
and
with
a
wait
time,
especially
the
wait
time
like
this,
even
if
it's
a
little
bit
slower
you're
still
going
to
be
a
lot
faster
running
on
the
knls,
but
perhaps
the
large.
If
there
was
a
much
smaller
number
of
large
jobs,
there
were
a
very
large
number
of
tiny
jobs.
Maybe
that
accounts
for
why
this
was
happening.
E
Yeah
that
so
I
think
those
our
numbers
appear
to
be
correct.
We
haven't
validated
them,
but
it's
true
that
the
wait
time
on
k
l
was
lower
in
july.
It's
it's
actually
back
up
again.
E
The
haswell
number
is
a
bit
suspect
because
of
such
a
big
increase,
one
of
our
main
problem
in
weight
in
measuring
wait
times
is
the
scheduler
does
not
give
us
an
easy
way
to
eliminate
jobs
that
were
submitted
that
are
not
eligible,
so
in
other
words,
one
project
might
submit
2
000
jobs
at
once
and
since
only
a
subset
of
them
are
eligible,
some
there's
some
limit
on
how
many
eligible
it
might
take
three
or
four
months
for
those
jobs
to
actually
drain
through
the
system,
because
they
can
only
run
so
many
at
a
time
and
that
really
skews
the
wait
time
once
they,
but
once
the
last
one
in
that
queue
finally
runs
in
december
or
whatever
it
had
a
wait
time
of
like
four
months
or
something
so
that's
another
possibility.
A
D
E
Is
true
that
that
the
wait
time
on
k
l
was
was
down,
but
now
it's
back
up
again.
A
Good,
okay,
all
right!
Thank
you
very
much
richard
all
right,
so
I
think
that's
everything
for
this
month.
Thank
you
so
much
for
attending.
If
there
are
any
follow-up
questions
or
any
last,
things
feel
free
to
put
them
in
the
slack
channel
or
I'm
still
here,
I'm
not
going
anywhere.
Quite
yet
and
yes,
thank
you
very
much
any
follow-up
questions
and,
if
not
have
a
good
lunch,
have
a
good
day.