High-Risk Technologies Applications to Fusion (and Accelerator) Physics Presenter: J.R. Cary †* † Tech-X Corporation; * University of Colorado October 7, 2008 --------------------- Session Chair: Phil Colella Preparatory remarks Risks for application developers Rewards in fusion and accelerators Promising technologies Horizons: general remarks Some project ideas
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Hig
h-R
isk
Tec
hnol
ogie
sA
pplic
atio
ns to
Fus
ion
(and
Acc
eler
ator
) P
hysi
cs
Pre
sent
er: J
.R. C
ary
†*
† Tec
h-X
Cor
pora
tion;
* U
nive
rsity
of C
olor
ado
Oct
ober
7, 2
008
----
----
----
----
----
-
Ses
sion
Cha
ir: P
hil C
olel
la
Pre
para
tory
rem
arks
Ris
ks fo
r ap
plic
atio
n de
velo
pers
Rew
ards
in fu
sion
and
acc
eler
ator
sP
rom
isin
g te
chno
logi
esH
oriz
ons:
gen
eral
rem
arks
Som
e pr
ojec
t ide
as
2
Ste
p 4:
Cat
alog
you
r tr
ansg
ress
ions
: I a
m a
n ou
tlier
•C++
for
scie
ntifi
c co
mpu
ting
sinc
e 95
•As
soon
as
I hea
rd a
bout
it, I
had
–S
ound
on
my
grou
p w
eb p
age
–A
non
linea
r dy
nam
ics
appl
et•R
e fu
sion
: rea
lly a
n ou
tlier
•I a
m a
n O
AS
CR
sym
path
izer
•But
my
brea
d an
d bu
tter
depe
nds
on d
oing
goo
d co
mpu
tatio
nal s
cien
ce
Pal
ahni
uk (
Cho
ke):
The
four
th s
tep
in th
e tw
elve
-ste
p pr
oces
s is
to k
eep
a re
cord
of y
our
addi
ctio
n, r
ecor
ding
all
your
tra
nsgr
essi
ons,
pas
t and
pre
sent
.
3
Who
sho
uld
driv
e to
ol d
evel
opm
ent?
Too
lers
(to
ol
deve
lope
rs)
or a
ppse
rs (
appl
icat
ion
deve
lope
rs)?
•T
echn
olog
y dr
iven
pro
duct
s ca
n m
iss
the
mar
k–
Exc
essi
ve d
evel
opm
ent c
ycle
(de
bug,
bui
ld, t
est)
–F
ive-
step
wor
kflo
ws
do n
ot r
equi
re c
ompl
ex m
anag
eme
nt–
Ove
rsim
plifi
catio
ns (d
istin
guis
hing
ser
vice
s on
bas
is o
f in
terf
ace)
•B
ut M
arke
t driv
en p
rodu
cts
are
ofte
n to
o la
te in
a
smal
l fie
ld
like
com
puta
tiona
l sci
ence
–C
ompu
tatio
nal s
cien
tists
hav
e to
com
e up
with
a s
olut
ion
to
surv
ive,
so
alre
ady
have
I/O
, com
pone
nt m
odel
, grid
s, d
ata
stru
ctur
es•
Dec
oupl
ed to
ol/a
pps
deve
lopm
ent b
est p
ract
ices
:–
Too
lers
res
pond
to n
eed,
pro
vide
som
ethi
ng b
ette
r–
Too
lers
pro
vide
mig
ratio
n pa
th–
App
sers
ado
pt o
n ne
xt c
ode
cycl
e (e
.g.,
mod
erni
zatio
n pr
ogra
m)
•T
oole
r opp
ortu
nitie
s be
st in
fiel
ds th
at a
re in
a
mod
erni
zatio
n pr
ogra
m o
r ha
ve a
his
tory
of m
oder
nizi
ng
4
Col
labo
ratio
n is
the
right
app
roac
h to
de
velo
ping
use
ful t
ools
•See
the
need
s of
app
licat
ion
deve
lope
rs fi
rst h
and
–F
ast I
/O: d
on't
give
up
met
adat
a•S
ee th
e dr
iver
s of
app
licat
ion
deve
lope
rs fi
rst h
and
–S
cien
ce!
Not
FLO
PS
, abs
trac
tion,
…•W
orki
ng w
ith a
pps
deve
lope
rs g
ive
appr
ecia
tion
for
hurd
les
(Why
are
thos
e ap
pser
s so
st
upid
as to
not
see
the
adva
ntag
es o
f my
tech
nolo
gy?)
•One
app
roac
h–
Bui
ld a
pplic
atio
n–
Ext
ract
goo
d id
eas
into
libr
arie
s–
Ref
acto
r ap
plic
atio
n to
use
libr
arie
s
For
sof
twar
e re
use,
this
latte
r is
our
app
roac
h at
Te
ch-X
5
Ris
ks: w
hy w
ould
an
appl
icat
ion
deve
lope
r ad
opt n
ew to
ols,
ev
en if
pai
d?
•Num
eric
al J
ava
•HP
C, H
PF
ortr
an•H
aske
ll, A
PL,
….
•The
Grid
?•S
CO
NS
•PO
OM
A
The
web
is li
ttere
d w
ith a
ppro
ache
s
The
com
puta
tiona
l too
l dev
elop
er m
akes
his
car
eer
on id
eas,
ev
en if
the
tool
is n
ot u
sed
The
com
puta
tiona
l sci
entis
t ris
ks h
is c
aree
r by
mak
ing
the
wro
ng to
ol c
hoic
e
Wha
t hav
e ap
ps d
evel
oper
s le
arne
d ab
out n
ew la
ngua
ges,
pr
omis
es o
f Nirv
ana,
…?
6
Typ
es o
f Ris
k
The
poi
nt o
f the
se p
roje
cts
is to
dev
elop
new
cap
abili
ties
that
, fro
m th
e po
int o
f vie
w
of th
e ap
plic
atio
ns c
omm
unity
, are
too
high
-ris
k to
be
carr
ied
out a
s bu
sine
ss a
s us
ual.
Cla
ssify
the
risks
of p
ossi
ble
proj
ects
in th
e fo
llow
ing
cate
gorie
s:
oW
ell-c
hara
cter
ized
app
licat
ion
of a
new
tech
nolo
gy –
risk
com
es fr
om u
sing
an
exot
ic m
etho
dolo
gy in
a p
rodu
ctio
n sc
ienc
e ap
plic
atio
n, re
quiri
ng h
arde
ned
impl
emen
tatio
ns o
f the
met
hodo
logy
and
a b
ridge
bet
wee
n th
e ap
ps d
omai
n an
d th
e ex
pert
s in
the
new
tech
nolo
gy. E
xam
ple:
impl
emen
ting
an e
xist
ing
mod
el in
a
new
pro
gram
min
g la
ngua
ge o
r pr
ogra
mm
ing
fram
ewor
k.
oW
ell-e
stab
lishe
d te
chno
logi
es a
pplie
d to
a n
ew p
robl
em a
rea
–ris
k co
mes
fr
om w
heth
er th
e m
etho
dolo
gy c
an b
e su
cces
sful
ly m
odifi
ed to
mee
t pro
blem
-sp
ecifi
c ne
eds.
Exa
mpl
e: A
MR
for
clim
ate.
oF
unda
men
tal n
ew a
ppro
ache
s, p
artic
ular
ly in
dom
ains
whe
re th
ere
is li
ttle
prio
r ar
t in
mod
elin
g. E
xam
ple:
Het
erog
eneo
us s
patia
l mod
elin
g in
cel
l bio
logy
.
oO
ther
ris
ks n
ot li
sted
her
e.
7
Com
puta
tiona
l app
licat
ion
deve
lope
rs p
refe
r th
e bo
ttom
of t
he s
oftw
are
risk
hier
arch
y•
Fai
lure
(e.
g., t
echn
olog
y dr
oppe
d by
sp
onso
r) r
equi
res
proj
ect d
o-ov
er–
Lang
uage
cha
nge
–G
ivin
g up
mai
n–
Ado
ptin
g an
othe
r's d
ata
stru
ctur
es•
Fai
lure
req
uire
s ex
tens
ive
code
rew
ritin
g–
Cod
e to
fund
amen
tally
diff
eren
t ap
proa
ch (
mes
sagi
ng -
> th
read
ing)
•F
ailu
re r
equi
res
min
imal
cod
e re
writ
ing.
–C
ode
to n
ew A
PI
•F
ailu
re r
equi
res
switc
hing
to a
new
ex
ecut
able
–R
epla
ce G
nuP
lotP
y w
ith m
atpl
otlib
•F
ailu
re r
equi
res
switc
hing
out
peo
ple
Bas
ili e
t al,
"Und
erst
andi
ng t
he H
igh-
Per
form
ance
Com
putin
g C
omm
unity
: A S
oftw
are
Eng
inee
r’s P
ersp
ectiv
e,"
IEE
E
Sof
twar
e, J
ul/A
ug,
29 (
2008
).S
hash
arin
a et
al,
"FA
CE
TS
–A
Mul
tiphy
sics
Par
alle
l C
ompo
nent
Fra
mew
ork,
" C
OM
PF
RA
ME
200
8 (K
arls
ruhe
, G
erm
any)
.
8
Fus
ion
appl
icat
ions
: to
pred
ict c
atas
trop
hic
mac
rosc
opic
mot
ion
or tu
rbul
ent t
rans
port
•Cor
e–
Mac
rosc
opic
mot
ions
, but
sta
rted
by
boun
dary
la
yer
inst
abili
ties
–M
icro
turb
ulen
ce d
eter
min
es tr
ansp
ort f
luxe
s–
Fas
t par
ticle
s•N
eutr
al b
eam
s•R
F g
ener
ated
•Fus
ion
prod
ucts
•Edg
e–
Mac
rosc
opic
mot
ions
, ELM
s–
Mod
erat
e tu
rbul
ence
–C
haot
ic fi
eld
lines
–A
tom
ic p
hysi
cs–
Wal
l int
erac
tion
We
got p
artic
les!
Impa
ct r
equi
res
ALL
of t
he p
hysi
cs
9
Rew
ards
ste
m fr
om s
cien
tific
dis
cove
ry (
from
th
ere,
gen
eral
cod
es)
Gen
eral
•D
isco
very
of a
new
m
echa
nism
•P
redi
ct th
e pa
ram
eter
s of
the
dom
inan
t m
echa
nism
•O
utlin
e a
serie
s of
ex
perim
ents
to
dete
rmin
e th
e do
min
ant
mec
hani
sm
Fus
ion
•N
eocl
assi
cal
tear
ing
mod
e st
abili
zatio
n an
d/or
dr
ive
to u
nsta
ble;
E
LM g
ener
atio
n by
pe
elin
g-ba
lloon
ing.
•N
orm
aliz
ed b
eta
as
indi
cato
r for
di
srup
tions
•S
how
ped
esta
l he
ight
dep
ends
on
som
e pa
ram
eter
co
mbi
natio
n
Acc
eler
ator
s•
Sel
f tra
ppin
g in
LW
FA; c
rab
cavi
ties
•W
hich
cav
ity h
as
the
grea
test
gr
adie
nt w
ithou
t qu
ench
ing?
Wha
t pu
lse
shap
e is
be
st fo
r se
lf tr
appi
ng?
•C
ompu
te o
ptim
al
cavi
ty, e
xper
imen
t do
es fi
nal
adju
stm
ent
10
Tech
nolo
gies
that
look
pro
mis
ing
(to
me)
Mul
tigrid
, AM
G, o
ther
sol
vers
rou
tine,
but
I ke
ep m
y da
tast
ruct
ures
. Now
dec
isio
ns b
eing
mad
e on
dev
elop
er
conv
enie
nce
(fle
xibl
e bu
ild, t
rue
seria
l, …
)•
Aut
otun
ing
(rea
lly n
eeds
em
bedd
ed C
S/A
mer
s =
tool
ers)
•M
assi
vely
par
alle
l on
chip
(G
PU
, Cel
l) fo
r bo
th fi
eld
s or
par
ticle
s•
Hig
her-
orde
r em
bedd
ed b
ound
arie
s (h
ow c
an w
e lib
rar
ify?)
•D
iver
genc
e pr
eser
ving
AD
I ele
ctro
mag
netic
s (t
oo s
peci
aliz
ed?)
•C
ompo
nent
con
cept
s co
uld
use
deve
lopm
ent w
ithin
app
licat
ions
, re
luct
ant t
o gi
ve u
p m
ain
•N
ew la
ngua
ges:
mus
t be
C(C
++)-
calla
ble,
usa
ble
in e
xist
ing
fram
ewor
ks
mor
e m
atur
atio
n ne
eded
•Lo
ad b
alan
cing
libr
arie
s: d
o I h
ave
to g
ive
up m
y d
ata
stru
ctur
es?
mor
e ris
ky
•V
isua
lizat
ion,
incl
udin
g ra
y tr
acin
gm
inim
ally
ris
ky
11
Hor
izon
: gen
eral
rem
arks
•F
or e
ach
proj
ect,
answ
er th
e fo
llow
ing
ques
tions
.o
Giv
e a
timel
ine
for
prog
ress
, ov
er a
3-5
yea
r tim
e sc
ale.
oW
hat w
ould
be
an o
ptim
al e
nd
stat
e in
10
year
s ?
oW
hat i
s th
e le
vel o
f effo
rt
requ
ired
to m
eet t
hese
goa
ls ?
oW
hat w
ould
be
the
orga
niza
tiona
l str
uctu
re o
f the
co
llabo
ratio
n be
twee
n th
e ap
ps
and
the
high
-ris
k m
ath
/ CS
te
am ?
oW
hat e
xter
nal d
epen
denc
ies
need
to b
e ta
ken
into
acc
ount
(e
.g. e
xist
ence
of s
uppo
rtin
g S
ciD
AC
infr
astr
uctu
re)?
•A
ny p
roje
ct h
as to
pro
duce
ph
ysic
s w
ithin
two
year
s.•
Afte
r 10
yea
rs, a
pro
duct
ion
code
for
both
gov
ernm
ent
and
indu
stry
•C
entr
al, i
nter
disc
iplin
ary
team
of 3
-5 w
ith
cont
ribut
ors
from
oth
er
inst
itutio
ns.
•C
entr
al te
am w
ith b
oth
apps
ers
and
tool
ers,
ad
ditio
nal r
emot
e te
ams
with
wel
l-def
ined
task
s•
Loca
l dev
elop
men
t en
viro
nmen
t.
12
Pro
ject
1: A
utot
uned
PIC
cod
es
•P
IC c
odes
rou
tinel
y ge
t bet
ter
perf
orm
ance
with
sor
ting
–P
artic
les
in c
ell [
0, 0
], th
en [0
, 1],
then
…
–T
imin
g te
lls w
hen
to s
ort
–B
ut p
artic
les
gath
er fi
elds
from
4
cells
–B
ette
r to
sor
t all
in c
ells
{[0
, 0],
[0, 1
], [1
, 0],
[1, 1
]} fo
llow
ed b
y ne
xt 4
and
so
on?
Gro
up b
y 9?
•P
artic
les
rout
inel
y gr
oupe
d–
Par
ticle
s co
me
and
go in
a s
imul
atio
n–
Mem
ory
man
agem
ent a
ided
by
use
of
rag
ged
arr
ays
of s
ome
max
imum
siz
e
Aut
otun
ing
coul
d he
lp u
s fig
ure
our
the
best
par
ame
ters
for
thes
e an
d ot
her
codi
ng c
hoic
es
RIS
KS
•In
vasi
ve c
odin
g m
ay
not b
e m
odul
ariz
able
RE
WA
RD
S•
Pro
duct
ivity
in
crea
ses
by fa
ctor
s of
10
on e
xist
ing
hard
war
e
13
Pro
ject
2: a
n au
totu
ned
mul
ti-m
ini-d
omai
ned
elec
trom
agne
tic P
IC c
ode
•Ver
y sm
all d
omai
ns
para
lleliz
e w
ell w
ith F
DT
D o
n la
rges
t har
dwar
e•I
f on
sam
e ch
ip, c
an th
ese
dom
ains
be
even
sm
alle
r?
Sm
alle
r eno
ugh
to fi
t in
cach
e?•T
wo-
leve
l ind
exin
g? Ski
nG
uard
Bod
y
RIS
KS
•In
vasi
ve c
odin
g m
ay
not b
e m
odul
ariz
able
RE
WA
RD
S•
Pro
duct
ivity
in
crea
ses
by fa
ctor
s of
10'
s on
mod
est
mul
ticor
e
14
Pro
ject
3: G
PU
or
Cel
l bas
ed c
ompu
tatio
ns o
f m
ulti-
min
i-dom
aine
d sy
stem
s
•See
pre
viou
s sl
ide
Mig
ratio
n pa
th:
Pro
ject
1, 2
, 3
RIS
KS
•In
vasi
ve c
odin
g m
ay
not b
e m
odul
ariz
able
RE
WA
RD
S•
Pro
duct
ivity
in
crea
ses
by fa
ctor
s of
100
's o
n m
assi
ve
mul
ticor
e
15
Pro
ject
4: H
ighe
r-or
der
embe
dded
bou
ndar
y co
mpu
tatio
ns
•Em
bedd
ed b
ound
ary
algo
rithm
s pr
ovid
e bl
azin
g sp
eed
with
ac
cura
cy•H
ow d
o w
e cr
eate
a
libra
ry?
RIS
KS
•A
sym
met
ric m
atric
es g
ive
long
-tim
e in
stab
ilitie
sR
EW
AR
DS
•In
crea
sed
accu
racy
, sp
eed
16
Pro
ject
4: G
PU
or
cell
base
d co
mpu
tatio
ns o
f E
M u
sing
AD
I
•A
DI a
ppro
ache
s ha
ve
natu
ral d
ata
brea
kup
•A
lway
s st
able
•T
rans
posi
tions
?
RIS
KS
•G
ener
aliz
atio
n to
hig
her
orde
r pro
blem
atic
?R
EW
AR
DS
•A
bsol
utel
y st
able
, tim
e st
eps
dete
rmin
ed b
y pr
oble
m n
ot n
umer
ics
17
Pro
ject
5: C
o-de
velo
pmen
t of c
oncu
rren
t pa
ralle
l com
pone
nt fr
amew
orks
•C
ompo
nent
in b
road
est s
ense
mea
ns o
nly
mod
ular
ity,
so
mot
herh
ood,
app
le p
ie, …
•W
hat d
oes
it m
ean
in p
hysi
cs?
–In
terf
aces
, but
ther
e is
som
ethi
ng m
ore:
two
com
pon
ents
m
eetin
g at
a s
urfa
ce p
rovi
de th
e sa
me
inte
rfac
e, b
ut
–Im
plem
enta
tions
: sim
ple
mod
els
to c
ompl
ex–
Inst
ance
s: m
ultip
le n
eutr
al b
eam
s•
Wha
t doe
s it
mea
n in
CS
?–
Run
time
disc
over
y•
Hig
h ris
k pr
ojec
t–
extr
act,
gene
ral s
oftw
are
from
an
appl
icat
ion
fram
ew
ork
–bu
ild a
n ap
plic
atio
n in
a n
ew a
rea
Una
bash
edly
FA
CE
TS
orie
nted
18
Pro
ject
s 6-
∞ ∞∞∞: M
ore
and
mor
e ris
k
•A
SC
R lo
ad b
alan
cing
so
ftwar
e? (
Giv
e up
dat
a st
ruct
ures
?)•
Mul
tiblo
ck fl
uid
solv
er fo
r ed
ge p
lasm
a? (N
eeds
be
tter
redu
ced
mod
els
in
the
edg.
)
RIS
KS
•G
ivin
g up
dat
a st
ruct
ures
…R
EW
AR
DS
•F
aste
r con
verg
ence
Nee
d m
ore
time
to fl
esh
out
19
No
conc
lusi
ons
this
is a
wo
rksh
op
!
Related Documents
