Folie 1 - SURFsara

tongueborborygmusΗλεκτρονική - Συσκευές

7 Νοε 2013 (πριν από 3 χρόνια και 5 μήνες)

51 εμφανίσεις

Beyond Simulation

New e
-
Science Perspectives on Industrial Innovation



Emile Aarts, Philips Research

1

1

2

Setting the scene

3

Where past and future meet

Source: H.R. Wieland, Computergeschichten, 2011

4

Mainframe computing: empowering organizations

5

Personal computing: empowering people

6

Mobile computing: any time and anywhere

7

Embedded computing: integration into our fabric

8

Computing at Philips

9

Philips pioneered with computing

1960:

PASCAL

Fastest computer in the world at that time



Memories: magnetic drums, magnetisch
rings, and B
-
memory


12,000 vacuum tubes, 10,000 transistors,
15.000 diodes


Power: 10 kW


Accuracy:

34 decimals


Special feature: music player

1981: P2000

Home computer



2.5 MHz Z80 microprocessor


4 (16) kB ROM, 16 (64) kB RAM en 2 kB video
RAM.


Rom
-
key cartridge and I/O
-
module


Built
-
in mini
-
cassettedrive


A single rom
-
key consists of four 2732
-
chips

10

Philips developed its own large
-
scale computing
facilities


Philips Computing Centre


IBM 380 Mainframe computer


Central facilitiy

Philips Computing Centre at the Bosch Dijk, Eindhoven, 1960

11

Magnetic properties

First principles calculation of magneto
-
crystalline anisotropy energy

Source: G. Daalderop, P. Kelly, M. Schuurmans 1983


Ab
-
initio calculation


Linear muffin
-
tin orbital method in
the atomic sphere approximation
using the local
-
spin density
approximation calculations


Convergence of the Brillouin
-
zone
integral of single
-
particle
eigenvalues


Comparison with full potential linear
augmented plane wave
approximation



12

Simulated Annealing

Source: E. Aarts and J. Korst, 1985

Monte
-
Carlo simulation

13

Trimedia


5 to 8 issue slots filled with up to 45
functional units


128 32
-
bit general purpose registers


SIMD & DSP operations


32
-
bit IEEE
-
754 floating point
operations


64 KB Instruction cache


28 KB Data cache


Separate memory and peripheral bus
interfaces


8 built
-
in timers


64 built
-
in vector interrupts


Supported by an ANSI compliant
C/C++compiler tool chain

Simulation of a VLIW Media Processor

Source: G. Slavenburg et.al, 1997

14

Philips has changed its portfolio over time

Philips has an inspiring mission



“We are a global company of leading
businesses creating value with
meaningful
innovations

that improve people‟s
health and
well
-
being
.”

15

15

15

15

OLED simulation


Light extraction into air




Simulations reveal that up to 80% of
the light emitted by the substrate can
be extracted by appropriate optical
surface structures



50% more lumen output is obtained
with high index (n= 1.8) structures
compared to low index structures
(n=1.5)



Simulations are based on ray
-
tracing
methods applying scattering grids
and matrices with numerical solutions
of the Maxwell equations


Micro
-
lens array

Pyramid array

16

Source: H. Greiner , 2007

17

Clinical Decision Support for DNA Sequencing


Huge Amounts of Data for Decision
Making: 1000 samples, 100GB/sample



Industrial grade Unix server with
terabytes of storage



Statistical learning



Accurate tumor sub
-
typing



Therapy response prediction for chemo
and biological therapy for cancer



Whole genome sequencing:
Predisposition (BRCA1, BRCA2, KRAS…
mutations, Lynch syndrome)

Diagnostic
-
level software that uses sequenced genome data to
correlate with other clinical data and aid in clinical decision making


Source: N. Dimitrova, 2011

18

Modeling of complex biological systems

understanding the human coagulation system


A full coagulation model involves
hundreds of protein interactions,
blood flow effects and a spatio
-
temporal simulation of clot formation



Optimal parameter ranges are
estimated via computation intensive
Markov chain Monte Carlo methods,
on a grid of parallel computers



We estimate the many model
parameters by fitting the model to
experimental data and theoretical
constraints, piece by piece

Source: B. Bakker, 2011

19

Recent trends and opportunities

20

Compute facilities usage at Philips Research

1980

1985

1990

1995

2000

2005

IBM Mainframe


VM/CMS
-

MVS

VAX/VMS

Apollo Workstations

GRID

HP
-
UX Central
Compute
Cluster
-

Linux

Silicon
Graphics

2010

21

BiG Grid Usage in 2011 @ HTC

bbmri.nl

1%

enmr.eu

39%

lsgrid

10%

Philips

50%

vlemed

0%

5.3 Million CPU Hours

bbmri.nl

0%

enmr.eu

47%

lsgrid

6%

Philips

46%

vlemed

1%

1.6 Million JOBS

In 2011, Philips Research executed 720,000 jobs

that used 2.6 million CPU hours.

22

The ballpark of e
-
science is changing


Application solutions will become ambient


Sensor networks will generate big real
-
time data streams


Backbones will provide on
-
line Intelligence


23

New propositions

24

Digital pathology

Networked high resolution scanning device


Automatic digitization of tissue samples


Resolution 0.25

m/pixel (Nyquist
-
sampling
for “40X”)


Data rate of 600MB/s


(15 mm/0.25

m)
2

X 24 bit RGB = 11 GB raw
data


Digital archive of cases


Peer to Peer discussion at remote locations


Search for textbook cases: now done with
Atlas


Computer Decision Support algorithms

Source: M. Gilles, 2011

25



Discovery of behavioral patterns in large
sets of data of individuals


Classification of activity patterns, habits,
detecting stressful moments, etc .


Providing lifestyle interventions strategies
tailored to (groups of) individuals


Simulating effects of lifestyle interventions
based on social influencing models


Data rate:


Activity: 4 bytes/minute = ~ 2MB year


Conversion: 4MB per user


60.000 users* ~ 240 GB year



Direct Life

Activity monitor that provides feedback and coaching on lifestyle
management

Source: A. van Halteren, 2011

3D accelerometer @ 20 Hz



26

Intelligent Street Lighting

Network of communicating context sensitive street poles



Network simulator Omnet++ on BiG Grid
computer cluster



Validated scalable, reliable, and responsive
network protocols for Outdoor Lighting
control for scenarios of up to 10k light poles



Simulations
generate ~200GB
d
ata
, while
performing iterative parameter evaluation

for
various
relevant combinations

under
multiple random generator seeds



Tailored lighting conditions, low
-
energy
consumption, and „simplified
commissioning, control and maintenance



Source: D.
Görgen

2011

27

Finally

In conclusion


e
-
Science plays a major role at Philips Innovation


Compute power and storage capacity will become ubiquitous: the Cloud
will develop into the one and single networked facility


Methods matter more than bandwidth


Consultancy will become increasingly important


Sara can play the role of gatekeeper



28

28