►
From YouTube: NUG meeting 15 April 2021
Description
Tips about developing models with Mathematica, avoiding OOMs when using MKL with dynamic linking, and our topic of the day was a preview of NERSC's annual report
A
So
a
little
bit
about
this
meeting,
the
format
is
a
very
interactive
format.
Please
participate.
I
think
we
have
a.
A
So
we're
currently
sitting
at
a
little
under
20
people,
which
I
think
is
enough-
that
we
can
just
yeah
speak
up
when
you've
got
something
to
contribute.
A
Yeah.
Also,
discussion
for
for
preference,
putting
discussion
in
the
nurse
user
slack
rather
than
the
zoom
slack
is
good,
because
then
the
discussion
continues
past
the
meeting
and
still
visible
afterwards,
and
for
this
we'll
mostly
use
the
webinars
channel,
hashtag
webinars
and
then
excuse
user
slack
sue
will
follow
our
normal
agenda
this
month,
which
is
start
out
with
a
win
of
the
month
and
followed
by
today.
I
learned
some
opportunities
to
talk
about
what
you've
achieved
and
learned
recently
a
few
announcements
and
calls
for
papers
topic
of
the
day.
A
I
seem
to
have
missed
filling
this
in
on
the
slides,
but
so
for
today
we're
going
to
take
a
a
little
bit
of
a
look
at
nusk's
annual
report
and
particularly
the
processes
and
how
we
do
it
and
what's
what's
interesting
and
useful
about
it,
particularly
from
a
nurse
users
perspective.
A
So,
for
our
first
regular
session,
section
we'll
begin
with
the
win
of
the
month,
actually
very
important.
First
off
everybody
can
in
fact
see
the
slides
and
hear
me
everything's
working.
A
I
saw
a
I
saw
a
thumbs
up
there,
so
that's
good
yes
great
to
you
so
win
of
the
month.
The
idea
of
this
segment
is
to
show
off
something
that
you've
achieved
or
shout
out
something
that
you
know
somebody
else
has
achieved.
So,
for
instance,
you
might
have
had
a
paper
accepted,
you
might
have
simply
solved
a
bug.
A
That's
been,
you
know,
challenging
for
a
little
while
you
might
have
a
significant
scientific
achievement
could
be
something
that's
also
a
candidate
for
a
high
impact
scientific
achievement
award
or
for
an
innovative
use
of
high
performance
computing
award
or
for
a
science
highlight
basically
yeah.
The
whole
range
we're
interested
in
sharing
small
wins
as
well
as
big
wins.
A
Showcase,
so
I
did
kick
something
off
with
most
of
both
my
wins
and-
and
you
know,
lessons
learned
in
the
last
couple
of
weeks
have
been
around
setting
up
a
update
of
the
spac
instance
for
nurse,
so
you
may
have
heard
of
spec
before
it's
the
supercomputing
supercomputer
package
management
software.
A
So
it's
a
tool
for
building
software,
essentially
building
yeah
third-party
software,
for
you
know
both
to
help
us
install
software,
that's
available
for
everybody,
but
we
also
are
trying
to
set
things
up
and
we
in
fact
have
a
spec
instance
available
at
the
moment
you
can
do
modular
veil
spac
and
that's
what
I'm
working
on
updating
and
you
can
use
spec
to
in
a
lot
of
cases
fairly
easily
install
third-party
software
in
your
own
home
directory
for
your
own
use,
and
it
gives
you
a
lot
of
flexibility
to
install
exactly
the
version
and
your
variant,
compiler
options,
and
so
on
that
you
need
for
your
software
and
there's
a
lot
of
built-in
recipes
that
you
know
can
simplify
the
process
of
getting
it
working.
A
So
a
lot
of
my
successes
in
the
last
couple
of
weeks
have
come
around
getting
steps
in
updating
us
back
to
work
and
getting
it
to
build
some
software
successfully,
and
you
know
in
the
next
few
days,
hopefully
we'll
have
a
an
updated
module
with
a
newer
version
that
has
some
significant
improvements.
B
Robert
all
right.
Well,
I
just
wanted
to
mention
that
I
was
able
to
use
corey
to
solve
a
problem
that
I've
been
working
on.
B
I
am
involved
in
the
development
of
being
beam
dynamics,
codes
that
are
essentially
codes
that
do
high
order,
perturbation
theory
around
a
reference
trajectory
and-
and
I
specifically
developed
the
algebraic
being
transport
codes
and
using
mathematica
on
nurse.
B
I
was
able
to
essentially
use
the
baker
hamburger
house
baker
campbell,
how
star
theorem
to
generate
formulas
that
are
needed
to
compose
two,
the
algebraic
maps
through
up
to
tenth
order,
and
then
I
had
mathematica
write
out
those
formulas,
and
I
coded
them
into
a
program
that
I
then
tested.
B
Using
using
corey
for
the
calculation
and
and
and
it
works,
so
I
just
wanted
to
point
out
how
corey,
having
all
these
features
available,
not
just
big
computer
power,
but
also
access
to
mathematica
right.
There
made
it
possible
for
me
to
generate
these
formulas,
write
them
and
then
have
mathematica
spit
out
the
program
spread
out
the
long
computer
code
and
paste
it
into
another
code
that
I
then
submitted
to
a
job
on
corey
and
it
all
worked.
So
so
that's.
A
Really
interesting,
if
I,
if
I'm
interpreting
right,
I'm
actually
hearing
kind
of
several,
what
would
you
call
it
yeah,
interesting
aspects
and
there's
there's
a
whole
sort
of
story
of
application
of
algorithm
development
here?
Is
that
correct,
so
so
you've
used
mathematica
by
the
sounds
of
it
to
prototype
a
new
algorithm
that
you're
developing
and
tenth
order?
That's
that's
hugely
precise
right.
B
I
actually
use
mathematica
to
write
the
formulas
that
the
formulas
themselves
at
each
order,
the
size
group,
the
number
of
terms
grows
dramatically
and
it
is
quite
doable
to
do
calculations
through
orders
of
a
few
four
or
five
by
hand.
But
beyond
that
it
would
be
a
incredibly
tedious
thing
to
do.
B
But
basically,
the
starting
point
for
this
was
an
online
document
written
by
alex
drought,
which
is
about
a
2700
page
document
on
the
algebraic
methods
and
non-linear
dynamics.
If
anyone's
interested,
I
could
I
could
post
the
link
to
it,
and
that
document
describes
the
steps
in
an
algorithm
to
to
calculate
the
concatenation
formulas
order
by
order
right
and
and
then
I
was
able
to
reuse
some
existing
mathematica
code
that
I
found
in
a
paper
on
the
right,
nested
cbh
formula,
and
so
I
didn't
exactly
use
mathematica
to
prototype.
B
B
Into
the
computer
code
I
was
developing
and
and
then
tested
it
on
corey.
B
Oh,
this
is
a.
This
is
a
nothing
job.
This
is
a
quick
job
because
I
mean
a
few
seconds,
but
these
formula
because
they
can
the
concatenator
acts
on
them.
Oh,
does
it
take
a
few
seconds
yeah
a
few
seconds.
The
concatenator
at
the
moment
is
not
the
computational
bottleneck,
but
when
you
are
actually
tracking
particles,
then,
if
you're
doing
a
multi-particle
simulation,
the
parallelization
is
over
the
particles,
the
concatenator
itself.
B
It
doesn't
take
very
long
unless
you
consider
a
few
seconds
a
long
time,
which
could
be
a
long
time
if
you
have
to
do
thousands
and
thousands
of
concatenations,
but
for
this
test
I
just
wanted
to
see
if
the
high
order
concatenator
was
working.
B
B
A
C
Mathematica,
since
I
usually
just
use
it
to
solve
equations
and
stuff
on
my
laptop,
but
not
really
to
do
heavy
lifting
like
that,
just
wondering
if
the
mathematical
kernel
is
parallel.
B
I
don't
know
the
answer
to
that
question,
although
for
my
purposes
that
was
not
such
an
issue
there,
there
was
a
time
over
the
past
week
or
two
when
I
did
run
mathematica
jobs
up
to
eight
hours,
because
I
didn't
realize
that
there
was
a
way
to
make
my
jobs
run
faster
and
if
you'd
asked
me,
then
I
would
have
said
I
sure
wish
it
would
run
faster.
But
now
I
can
generate
the
the
tenth
order
formulas
with
mathematica
running
in
a
minute.
So
I
don't
know.
B
A
B
I
I
just
I
don't
happen
to
have
mathematica
on
my
laptop.
I
have
an
app
on
one
of
my
ancient
laptops,
but
not
on
the
laptop
I'm
using
now,
and
so
it
just
made.
It
was
very
convenient
to
log
in
to
ner
to
nurse.
Do
module
load
mathematica,
generate
my
formulas
and
paste
them
into
the
code
right
there.
So
and
these
things
and
these
formulas
are
hundreds
of
lines
long
at
their
worst
and
the
few
lines
long
at
the
best.
D
A
A
Yeah,
I
think,
if
you're
getting
to
you
know
where,
where
you're
using
mathematica
for
that
kind
of
heavy
computational
lift
running
it
running
on
you
know,
through
the
batch
system,
would
would
certainly
be
better
because
it's
a
you
know
a
little
bit
too
much
for
a
a
shared
login
node.
A
So
for
up
to
four
hours,
we've
got
the
interactive
queue,
which
is
a
reasonably
fast
way
of
kind
of
you
know
getting
onto
a
compute
node.
You
would
probably
need
to
set
things
up
a
little
bit
using
your,
for
instance,
nx,
so
that
you've
got
a
decent,
gui
connection.
If
you're
using
the
you
know
the
gui
for
it
but
yeah,
you
can
also
run
interactively
just
through
the
regular
queue
it
just
can
take
longer
for
the
job
to
get
in
and
be
a
little
bit
more
unpredictable.
A
You
know
in
terms
of
when
you
need
to
be
there
to
to
start
it
off
if
there's
an
interactive
component
but
yeah,
I
think,
there's
a
degree
of
you
know
sensible
judgment
there
right
now.
You
know
yeah
by
all
means
for
development,
work
and,
and
so
on,
like
that,
that's
a
that's!
A
great
use
of
the
login
nodes
as
yeah
yeah
as
the
intensity
starts
to
go
up,
and,
and
actually,
if
you
have
sort
of
your
questions
along
the
way,
how
do
I
make
good
use
of
the
compute
nodes?
A
A
A
That
was
really
good,
so
going
on
into
the
flip
side
of
the
question
is
the
the
today
I
learned,
and
this
is
kind
of
an
opportunity
to
you-
know.
A
And
learn
from
the
things
that
didn't
go
so
smoothly.
There's
this
old
saying
every
once
that
I
really
liked,
which
is
that
you
know
knowledge
comes
from
reading
the
fine
print
and
wisdom
comes
essentially
from
not
reading
the
fine
print
and
and
yeah.
This
in
a
way,
is
kind
of
what
we're
talking
about
here,
things
that
you've
discovered
it
doesn't
have
to
be
the
hard
way
it
can
be.
A
You
know
an
interesting
presentation
that
you
saw
that
you
know
you
think
is
worth
you
know,
calling
out
and
telling
other
other
nurse
users
about
yeah.
Does
anybody
have
an
interesting
story
of
something
they
learned
that
they'd
like
to
pass
on
a
lesson.
A
A
Cherry
picking,
specific
pull
requests
to
you
know
pull
just
a
particular
change
into
you
know
an
earlier
tagged
version
to
get
yeah
something
to
work
without
actually
using
the
cutting
edge
development
version
of
everything,
and
fortunately
it
wasn't
too
hard.
Although
I
did
manage
to
you,
know
mess
it
up
the
first
time
I
have
to
go
back
in
and
carefully
pull
out
just
the
pull
request
files
that
I
was
interested
in.
E
So
what's
happened
is
yeah,
so
I
actually
a
long
time
ago
when
we
switched
the
linking
default
on
corey
as
a
dynamic.
E
E
E
For
you
know
the
the
mcdrum,
I'm
sorry,
let
me
start
off,
so
this
environment
variable
is,
is
used
to
limit
the
mcd-run
memory
for
mkl.
So
if
we
don't
use
this,
then
it
will
just
use
its
unlimited
intel.
Mkl
can
use
whatever
available
just
to
use
unlimited
memory,
h,
mcd-run
memory,
so
after
you
know
set
to
this,
the
oom
problem
is
resolved.
E
So
actually
this
has
been
bothering
me
like
many
months
and
especially
we
need
to
use
dynamic
linking
for
checkpoint
restart
tool.
So
this
was
really.
You
know,
bothered
me
for
a
long
time,
but
now
I
think,
with
this
workaround,
the
problem
is
resolved,
so
I
think
if
any
users
who
whoever
use
mkl
on
knl
and
if
you
run
into
oom,
probably
you
consider
to
set
this
environmental
yeah
variable
it.
A
E
I
think
so
the
cache
mode
is
yummy
users
that
don't
have
the
control
right,
but
mkl
trying
to
use
this
mankind
to
you,
know,
control
the
memory
used
in
the
mcd
run
and
then
somehow
it
I
don't
know
for
whatever
reason
it
doesn't
work,
I
think
from
the
beginning.
I
guess
so.
Actually
this
environmental
module
is
something
I
I
mean
this
environment
variable
is
something
I
adopted
from
my
early
intel.
You
know
collaborators
yeah.
A
So
yeah,
that's
a
that's
a
good
tip
and
the
and
the
symptom
that
we
should
watch
for
is
if
you're
seeing
out
of
memory
oom
crashes
in
the
job,
particularly
with
dynamic
linking
and
on
knl
and
you're,
using
mkl.
A
A
I
see
in
the
chat
robert's
put
a
link
to
the
book.
This
is
the
2700
page
book.
B
A
B
Know
if
it's
getting
money
it's
getting
longer
by
the
day,
I
think
there's
a
lot
of
great
good
material
in
there
great
material
in
there
just.
A
See
anybody
else
have
a
lesson
or
tip
they'd
like
to
share.
A
A
We
have
the
what
used
to
be
called
quarterly
allocation
reductions-
and
I
don't
know
if
they're
specifically
quarterly
anymore,
but
there's
a
couple
of
allocation
reduction,
dates
and
they're
set
for
may
4
and
september
7.,
and
what
these
are
is
because
you
know
predicting
exactly
how
much
you're
going
to
need.
You
know
several
months
in
advance
is
not
necessarily
easy,
and
so
we
kind
of
you
know
monitor
the
the
burn
rate.
A
E
Sorry
to
interrupt
I'm
not
seeing
the
slides
advance.
A
Oh,
I
have
to
watch
that
a
little
bit
more
closely
thanks,
lisa
yeah.
So
if
you
have
any
sort
of
questions
about
the
allocation
reductions
and
and
the
reductions
that
go
back
into
the
pool
so
that
the
program
managers
are
able
to,
you
know
reallocate
the
you
know
the
unused
time
to
your
other
projects
that
need
it.
A
I
guess
there
are
some
circumstances
where
perhaps
you
you
know
that
your
project
is
going
to
use
most
of
its
allocation
during
a
very
short
time
of
the
year
and
it
might
be
after
may
4..
So
you
know,
if
you
do
have
these
known
circumstances,
you
can
you're
getting
contact
with
us
via
a
ticket
to
talk
about
it.
A
Yeah.
I
guess
that's
just
a
heads
up
with
those
allocate
those
dates
are
set.
We
have
a
couple
of
training
events
coming
up
on
may
7,
there's
a
training
event
for
the
mana,
which
is
mpi
agnostic.
A
I
forgot
more
than
the
next
challenge
stands
for
zinji
talk
a
little
more
about
this
network
agnostic,
transparent,
checkpointing
tool.
Do
you
want
to
say
anything
else
about
that
zinc.
E
A
Sounds
good
yeah,
so
if
your
application
is,
is
a
minimal
to
being
checkpointed
and
manner,
for
I
think
a
lot
of
applications
will
yeah
we'll
sort
of
do
it
automatically.
This
can
be
a
really
good
way
to
make.
I
guess
better,
tighter
use
of
gaps
in
the
queue.
A
We
also
have
a
tutorial
coming
up
very
soon,
actually
in
april
so
next
week
for
time
memory,
I
think
that's
how
you
pronounce
it
and
that's
a
a
tool
that's
been
developed
within
nurse.
I
don't
know
if
jonathan's
on
the
call
at
the
moment,
I
think
he
is
but
so
this
is
a
tool
for
your
analysis
of
the
you
know,
performance
and
memory
use
of
your
application,
and
so
the
tutorial
is
part
of
part
of
the
ect,
the
ecb
program
for
that
coming
up.
A
Normally,
this
meeting
happens
the
day
after
the
scheduled
maintenance,
but
just
because
of
you
know,
when
tuesdays
and
thursdays
and
wednesdays
happen
this
month,
the
schedule
maintenance
for
corey
is
next
wednesday.
A
It's
up
to
that,
and
you've
probably
already
already
noticed
that
the
hpss
archive
system
is
currently
in
a
maintenance
which
is
soon
to
finish.
A
A
Losing
data
to
purging
so
before
we
go
into
topic
of
the
day,
is
there
any
other
announcements
or
cfps
that
anybody
here
would
like
to
make.
A
A
So
anybody
who's
been
a
nurse
user.
For
you
know,
more
than
just
year
will
have
seen
nurse's
annual
survey.
Anybody
who's
submitted
tickets
will
have
seen
the
post-ticket
surveys.
We
make
these
sort
of
calls
regularly
for
nominations
for
awards
and
science
highlights,
and
you
know,
maybe
you've
wondered
what
does
nurse
actually
want
all
this
stuff
for
so
for
today,
we're
going
to
take
a
bit
of
a
sneak
peek
at
nurse's
annual
report.
A
Basically,
why
we
do
it?
Why
we
call
for
these
things
why
we
write
a
report,
how
we
use
the
various
data
that
we
gather
and
perhaps
most
importantly,
how
you
can
use?
I
guess
this
information
and
and
your
knowledge
of
what
nurse
kind
of
you
know,
goals
and
aims
are
to
improve
nurse's
ability
to
support
your
needs.
A
Immediate
stakeholder
is
the
doe
office
of
science,
so
the
good
news
is
that
what
the
doe
wants
from
nurse
is
for
us
to
be
able
to
demonstrate
that
what
nurse
provides
is
helping
its
users
to
succeed
in
furthering
doe
research
goals,
which
roughly
translates
as
we
win
when
you
win,
and
a
significant
component
of
this
is
the
annual
report
that
we
submit
to
doe
every
year,
where
we
basically
analyze
and
write
about
what
happened
at
nursk
and
the
know,
services
and
resources
that
nurse
provides
and
outcomes
from
things
like
the
user
survey.
A
A
So
what
the
doa
wants
from
nurse
is
for
us
to
demonstrate
that
the
facilities
and
resources
we're
providing
are
effective
for
our
users
to
advance
advance.
Your
research
and
we've
kind
of
got
broadly
speaking,
two
tools
for
doing
this.
One
is
to
find
out
what
our
users
have
achieved
and
showcase
it
and
the
other
is
find
out
what
impedes
our
users
from
making
progress
or
what
makes
it
difficult
and
yeah
identify
things
that
we
can
improve
to
cause
more
success.
A
So
showcasing
user
achievements,
one
kind
of
yeah
almost
almost
blunt
instrument
here,
but
that's
an
important
one-
is
simply
counting
publications.
So
you
know
more
public
research
is
more
evidence
that
does
investment
in
nurse
is
reaping
dividends,
yeah
with
the
subtext
thereof.
So
it's
worth
investing
more
yeah.
This
comes
up
about
fairly
early
in
the
report.
A
In
2020,
we
identified
a
little
over
1800
articles
like
1830
or
860
articles
that
cited
nursk
as
a
resource
that
was
used
or
for
support
in
their
research,
and
so
this,
of
course,
bounces
into
you
know
how
you
can
help,
which
is
publish
stuff.
That's
that's
great
for
you,
and
it's
also
great
for
us,
especially
if
you
can
add
some
citation
that
acknowledges
use
of
nurse
resources
in
there.
So
when
we
do
the
report,
you
know
we
essentially.
A
A
So
this
was
also
kind
of
a
a
little
bit
of
a
win-win,
because
by
you're
showcasing
particular
research.
That
also
you
know,
brings
the
successes
of
that
of
that
research
project
into
the
does
sites
as
well
and
yeah.
It
was
also
kind
of
a
nice
point
to
be
able
to
show
off,
for
you
know,
collaborators
and
potential
collaborators
and.
A
There's
a
bunch
of
links
down
there,
including
some
links
to
these
science
highlights
and,
and
these
tend
to
be
formatted
as
a
slide.
You
know
a
reasonably
predictable
format.
We
have
an
example
of
it
here,
for
instance
with
a
scientific
achievement.
What
the
significance
is
and.
A
Of
the
research
and
links
to
you
know,
who
did
it?
Who
was
involved?
There's
a
few
other
things
on
that
page
as
well.
So
there's
there's
some
sort
of
vignettes,
which
are
slightly
more
detailed
in
the
science
highlights.
Some
projects
also
from
this
page,
are
links
to
where
you
can
nominate
particular
achievements.
A
So
a
nurse
gives
out
a
couple
of
achievement
awards
for
high
impact
science
and
for
innovative
use
of
high
performance
computing,
and
this
is
you
know
another
way
for
both
us
to
demonstrate
to
the
doe,
and
I
guess
the
wider
public
that
yeah
this
is
a
valuable
resource
and
for
you
to
be
able
to
demonstrate
that
you're
making
good
use
of
it,
and
you
know
the
the
successes
of
your
own
work.
A
So
how
you
can
help
here
is
nominate,
something
you
can
you
can
self-nominate
or
nominate
work
that
you
know
that
somebody
else
a
colleague
or
you
know
another
project
you've
encountered
has
done
and
we
have
a
link
down
here,
which
is
all
under
this
science
tab
on
the
nurse
webpage
for
the
nomination
form.
A
So
then,
on
the
other
side
of
it,
we
have
seeking
and
acting
on
user
feedback,
and
this
is
where
we
come
into
all
the
different
surveys
and
yeah
so
forth
that
we
spread
around
so
nist's
annual
survey.
You
will
have
seen
we've
been
doing
it
for
a
long
time,
which
means
we've
built
up
quite
a
history
of
data,
and
it's
changed
a
few
times
over
that
time.
A
A
A
So
in
nurse
annual
survey?
It's
it's
gratifying
to
see
that
users
are
have
consistently
over
time,
reported
quite
high
satisfaction.
So
what
these
two
lines
here
are.
This
is
a
historical
graph.
It's
a
little
bit
too
small
to
see
very
easily.
It
goes
back
to.
A
I
think
it
goes
back
something
in
the
order
of
20
years
and
yeah.
We
have
a
target
which
is
roughly
corresponds
to
you
know,
moderately
satisfied
or
above
and
the
yeah
average
score
over
the
years
has
been
consistently
higher
than
it
one
change
that
happened
in
the
last
couple
of
years
as
we
we
switched
from
a
seven
point
scale
to
a
six
point
scale-
and
this
is
all
you
know.
E
A
See
that
in
the
survey
questions
or
in
the
survey,
when
you
call
it
solicitation
and
they're,
you
know
the
experts
on
doing
surveys
and
how
to
design
the
survey
to
you
know
make
sure
that
we're
getting
you
know
valid
and
useful
answers,
and
you
know
that
it's
a,
I
guess,
a
reasonable
and
hopefully
not
too
large
time
commitment
from
users
to
fill
out
so
and
there
are
different
categories
that
the
results
get
broken
down
into,
and
it's
also
sort
of
gratifying
to
see
that
the
average
scores
that
we're
we're
seeing
reported
consistently
quite
high.
A
Actually
so
so.
Actually,
we
would
like
to
thank
our
users
for
being
very.
You
know
generous
and
helpful
in
your
feedback
with
that.
A
The
other
aspect
of
the
survey
is
the
text
questions,
and
this
is
a
little
small
to
read,
but
so
we
ask
a
few
text
questions
about.
You
know
what
does
nurse
do
well
and,
and
what
can
nurse
do
to
serve
you
better
and
we
do
a
mixture
of,
or
rather
you're,
with
the
help
of
the
survey
experts?
We
do
a
mixture
of
analysis
of
both
the
the
scoring
sort
of
questions
and
the
text
questions
to
try
to
understand.
A
Basically,
you
know
what
is
working
and
we
should
keep
on
doing
and
what
is
not
working
or
could
use
improvements,
and
so
you
know
both
the
scores
and
the
text
is
really
helpful
for
this.
The
the
text
is
good
because
you
know
that
is
an
opportunity
for
you
to
describe
in
in
more
detail.
A
You
know
what
you
know,
works
or
doesn't
work,
and
why-
and
you
may
find
in
in
the
coming
weeks,
possibly
even
in
recent
weeks-
that
nurse
staff
might
contact
you
about
things
that
you've
said
on
the
survey
yeah
to
try
to
understand
better.
You
know
what
it
is
that
we
can
work
on.
A
So
how
you
can
help
is
fairly
straightforward.
Here
is
participating
on
participate
in
the
survey,
and
you
know
when
something
works
or
doesn't
work
yeah
describing
it
in
the
text
is,
is
actually
quite
helpful
to
us.
A
A
So
in
this
analysis,
yeah
we
dive
down
into
the
factors
that,
according
to
the
the
statistics
from
the
scores
and
from
the
analysis
of
the
text,
you
know
pain,
points
and
and
things
that
we
would
like
to
focus
on
in
the
next
year
and
so
for
2020.
A
A
There
are,
we
have
a
lot
of
nurse
users
doing
a
lot
of
projects,
a
finite
number
of
resources,
and
so,
of
course,
you
know
there
is
a
queue
and
it
can
be
long.
The
approaches
that
nurse
has
been
taking
to
try
to
you
know
alleviate
some
of
this,
but
part
of
it
is,
you
know,
buying
for
ever
more
resources,
but
the
other
part
is
tuning
and
tweaking
things.
So
we
make
adjustments.
A
For
instance,
to
you,
know
cue
settings,
q
policies
to
try
to
increase
the
system
utilization
because
you
know
cycles
where
a
cd
compute
node
is
sitting
idle
that
doesn't
help
anybody,
so
the
better
utilization.
Overall,
we
can
get
that
the
better.
That
means
that
jobs
are
getting
through
the
queue.
A
So
what
else
can
you
do
then
back
to
showcase
your
work?
You
know
that
helps
to
make
a
case
for
more
money
and
and
bigger
resources,
but
also
in
their
everyday
day-to-day
as
a
general
tip.
Shorter,
wider
jobs
tend
to
work
better
with
the
scheduler,
and
if
your
work
is
amenable
to
it-
and
this
is
where
the
transparent
checkpoint
can
can
really
help
as
well.
Take
a
look
at
flex,
jobs
and
using
checkpoint
e
to
work
with
a
scheduler
and
use
the
gaps
in
the
schedule,
and
this
increases
the
utilization.
A
A
Overall,
summary
might
be
that
nursk
has
some
really
detailed
documentation,
but
it's
quite
complex
and
it
can
be
challenging,
especially
for
new
users
to
you
know,
get
up
and
started
so
that
that
was
useful
for
us
to
discover,
and
it
was
part
of
the
the
motivation
behind
you
may
have
noticed.
We
have
consulting
appointments
and
one
of
the
consulting
appointments
is
nurse
101.
A
These
are
all
under
there's
a
link
down
here.
Actually,
on
help.nurse.gov
down,
on
the
left
hand
side,
you
can
book
a
consulting
appointment
and
there's
a
list
of
types
of
appointments
so
for
new
users,
the
nurse
101
might
be
a
good
one.
So
you
know
one
to
one
getting
started:
we're
also
working
to
improve
the
documentation,
particularly
with
new
users
in
mind,
and
here's
where
our
existing
users
can
join
the
effort.
A
A
I
don't
have
to
check
that,
but
on
docs.nurse.gov,
if
you
look
up
in
the
top
right
hand,
corner
kind
of
where
this
little
red
mark
here
there's
a
link
to
the
gitlab
repository
of
the
docs
source.
A
The
other
thing
that
users
called
out
as
being
a
pain
point
was
down
times,
and
this
is
somewhat
understandable
is
disruptive
when
the
machine's
unavailable,
especially
at
the
moment
you
know,
since
edison
retired
and
while
we're
still
preparing
pearl
mudder,
we
don't
have
a
second
system
which
means
that
when
there's
a
scheduled
maintenance
on
corey,
there's,
not
really
a
plan
b
available,
so
we're
anticipating
that
things
will
improve
in
the
near
future
here
with
pearl
mata,
partly
for
the
simple
reason
that
there
will
be
two
systems,
and
so
you
know,
if
one
of
them
is,
you
know
if
corey
is
down
for
maintenance,
pearl
matter
will
still
be
available
and
the
other
is
yeah.
A
Perlmutter
is
some.
Has
some
significant
differences?
I
guess
in
its
operating
system
it's
a
shasta
operating
system,
so
it's
kind
of
a
new
approach
from
craig
and
one
outcome
of
this
is
that
we
expect
the
maintenances
to
be
a
lot
less
disruptive.
A
So
in
terms
of
what
you
can
do
here,
probably
being
ready
for
pearl
mata
is
the
you
know
the
biggest
and
most
obvious
thing:
we
have
a
bunch
of
docs,
particularly
so,
if
you're
developing
code
and
looking
at
performance
under
performance
here
in
the
docs
there's
a
perlmutter
readiness
page,
which
has
lots
of
tips,
you
can
also
try
things
out
early.
A
We
have
a
small
development
system
with
a
handful
of
nodes
that
have
gpus,
which
are
not
exactly
what
perlmutter
will
have,
but
it's
sort
of
in
that
direction.
So,
if
you're,
you
know,
testing
out
your
workflows
or
developing
code
using
corey's
gpu
development
nodes
is
a
good
way
to.
You
know,
get
a
bit
of
a
feel
of
it
there
and
build
it
there
and
you
you
can
request
access
to
those
also
via
the
help
system.
A
So
I
think
from
a
those
are
the
things
that
I
guess
jumped
out
to
me
as
being
yeah,
particularly
interesting
things,
for
you
know
I'll
use
a
community,
and
you
know
things
that
perhaps
you
can,
you
know,
take
some
sort
of
action
about,
and
there
are
no
doubt
many
other
things
that
are
interesting
there.
So
other
parts
of
the
annual
report
go
through
some
of
our
operational
numbers,
so
yeah
we
kind
of
bounce
off
this
a
little
bit
for
the
last
month.
A
Numbers
section
of
this
meeting,
which
is
coming
up
actually
in
a
moment,
there's
a
section
of
the
report
where
we
talk
about
innovations
that
are
happening
at
nursk
to
improve
nurse
operations
and
also
usability.
So
this
is
this
is
almost
kind
of
you
know:
research
into
or
yeah
yeah,
research
and
development
intended
to
improve
user
experience
in
coming
years.
A
So
it's
it's
quite
a
long
report,
but
you
know
there's
some
interesting
stuff
there
to
read
and
it
will
soon
be
available
at
this
web
page.
Ultimately,
I
think
that
the
key
thing
here
is
that
you
know,
with
with
the
annual
report
being
a
fairly
significant
part
of
it.
Nurse
goes
reviewed
based
on
how
well
we're
able
to
support
the
success
of
our
users
so
yeah.
A
You
know
nominating
work
and
citing
nurse
in
publications
and
also
letting
us
know
in
surveys,
and
you
know
where
there
are
other
forums,
but
particularly
surveys,
because
that
gives
us
a
you
know,
a
body
of
work
that
we
can
analyze
and
you
know
sort
of
focus
on
a
bit
and
let
us
know
what
works
well
and
what
poses
difficulties
and
that
helps
us
to
direct
our
efforts
to
improve
support
kind
of
in
coming
years.
A
Here's
what
the
front
page
looks
like.
There
are
quite
a
few
pages.
What
have
we
got
122
pages,
so
we
probably
don't
have
time
to
go
into
too
many
details
of
the
actual
content
in
this
discussion,
but
we
do
have
a
few
minutes
for
if
anybody
has
any
questions
or
comments,
I'd
like
to
make
you
know
about
the
process
or
you
know,
I
guess
things
of
interest
in
the
report.
A
In
the
meantime,
though,
yeah
do
go,
take
a
look
at
this
link,
and
you
know
where
I
flick
through
the
2019
report
might
be
a
good
way
of
getting
an
overview
of
the
sort
of
things
we
asked.
Oh,
I
see
there's
a
couple.
A
D
A
Right,
yes,
I
think
we
do
have
that
same.
I
guess
we
call
it
factor
almost
almost
limitation
in
the
way
we
discover
these
papers
is
by
searching
for
citations.
A
A
We
have
a
couple
of
sections
left
coming
up,
we're
always
looking
for
topics
either
either
requests
or
suggestions
for
a
topic
of
the
day
if
you've
got
some
work,
you'd
like
to
showcase,
so
something
that
I
would
really
like
to
see
is
some
of
some
more
of
these
topic
of
the
days
being
nurse
users
showing
off
what
what
you've
achieved,
and
so,
if
you
have
something
that
you're
interested
in
doing
at
10
or
15
minutes,
essentially
lightning
talk
about
the
work
that
you're
doing,
even
if
it's
work
in
progress,
the
work
that
you're
doing
and
how
you're
using
nurse
resources
with,
maybe
some
you
know,
tips
and
ideas
that
other
users
can
bounce
off
about
the
things
that
you've
found
work.
A
I
see
because
she's
posted
in
the
chat.
Actually
this
is
this
is
a
good
idea
how
to
edit
and
contribute
to
this
documentation
as
a
monthly
topic.
That
would
be
a
good
topic.
Actually,
we
can
do
a
walkthrough
of
the
the
system,
so
we
use
a
system
called
mcdocs,
mk
docs
and
it's
essentially
marked
down
formatted
documentation
in
a
git
repository
that
we
accept
pull
requests
for
so
yeah
that
that
would
be
a
good
topic,
a
walk
through
how
to
contribute
thanks
kg.
A
And
finally,
last
month's
numbers,
so
much
uptime
wise,
we
had
a
a
really
good
month.
We
had
no
unscheduled
down
times.
I
think
the
message
of
the
day
did
show
one
unscheduled
downtime,
but
it
turned
out
to
be
a
false
alarm
triggered
on
utilization
dropped,
and
it
was
simply
that
a
very
large
job
was
getting
ready
to
start
and
a
lot
of
nodes
had
to
be
cleared.
So
the
only
downtime
that
we
had
was
the
scheduled
maintenance.
A
The
utilization,
so
the
the
large
jobs
was
actually
up
a
little
bit
from
normal.
Our
target
is,
I
think,
25
of
the
workload
being
large
jobs,
so
work
that
can't
progress
without
a
really
large
scale
system
such
as
quarry
utilization
was
high,
but
you
know
in
in
recent
months
it's
actually
been
sitting
up
around.
The
kind
of
you
know,
94
95,
and
I
kind
of
wonder
if
part
of
the
relative
dip
is
because
you
know
large
jobs
can
have
gaps
scheduling
around,
but
this
is
where
this
is
partly.
A
Our
official
definition
of
large,
I
think,
is
1024
or
more
nodes.
Okay,
so,
and
I
think
we
particularly
only
consider
jobs
on
the
k
l
nodes
for
that,
because
there
are
only
about
two,
a
little
over
2
000
as
well
nodes,
so
the
kind
of
half
of
the
machine.
I
might
need
to
to
verify
that.
A
I
think
we
do
have
it
in
the
policies
written
on
the
in
the
documentation
but
yeah,
so
so,
1024
job
nodes
plus
counts
as
a
large
job
and
also
becomes
eligible
for
a
discount
in
the
job
cost.
A
New
tickets
yeah
a
chart
of
how
these
tickets
move
over
over
time
would
probably
be
interesting.
The
numbers
do
tend
to
be
reasonably
consistent.
You
know
so
yeah,
five
to
seven
hundred
a
month,
newty
gets
closed
to
get
our
current
backlog
as
of
a
couple
of
weeks
ago
was
a
little
over
500
tickets
flash
and
that's
all
we
have
for
today,
and
it's
right
at
the
top
of
the
hour.