Digital Publishing,

High Performance Computing,

and the Grid

Jan Allebach
allebach@purdue.edu
Electrical and Computer Engineering
Purdue University

Gelato Strategy Council Meeting 24-26 May 2004

 Synopsis
z Evolution of digital publishing
z Why is digital publishing important?
z What can high performance computing and the grid offer?
z A use case for digital publishing on the grid

Gelato Strategy Council Meeting 24-26 May 2004

 Digital publishing is the result
of three distinct threads of
technology development

z Printing
z Digital computers
z Desktop publishing

Gelato Strategy Council Meeting 24-26 May 2004

 Printing requires ink, paper,
and a means to form the image
z The use of a colorant for writing arose with the use of papyrus
scrolls for writing in 2400 BC in the Middle East.
z The Chinese developed paper from macerated hemp fibers in
water suspension in 150 BC.
z The first book the Diamond Sutra was printed using wood-block
in China in 868.

Gelato Strategy Council Meeting 24-26 May 2004

 Further developments made
printing more efficient
z A papermaking mill was established in
Capellades, Catalonia in 1238.
z Gutenberg invented press and metal type
in 1450 precipitating a media “explosion”
in Europe – by 1500 more than 9 million
printed books were in circulation.
z Lithography was invented by Senefelder
in 1799.
z Principle of offset press using a rubber
transfer blanket was developed in early
1900s.

Gelato Strategy Council Meeting 24-26 May 2004

 The history of electronic
computing is much more recent
z The 30-ton ENIAC computer with 19,000 vacuum tubes
and 1,500 relays was placed in service in 1945.
z IBM introduced the 5150 PC in 1981.

Gelato Strategy Council Meeting 24-26 May 2004

Desktop publishing required a low-cost
printer and computer with a GUI, and a
simple page description language
z Apple introduced Macintosh PC in 1984.
z HP introduced LaserJet printer in 1984.
z Adobe introduced Postscript in 1985.
z Aldus developed Pagemaker in 1985.

Gelato Strategy Council Meeting 24-26 May 2004

 Commercial presses are based on
“impact” printing technologies
z Letterpress and flexography
z Offset lithography
z Gravure
z Intaglio

Heidelberg Speedmaster SM 74 offset press

20”x29”, 2-color, 10K sheets/hr.

Gelato Strategy Council Meeting 24-26 May 2004

 Desktop printers use “non-
impact” technologies

z Inkjet
z Electrophotography

Laser scanning mechanism

Schematic of laser (EP) printer

Gelato Strategy Council Meeting 24-26 May 2004

 Digital publishing is like desktop
publishing on steroids
z Desktop publishing software for document preparation.
z Scaled-up desktop printing technologies

HP Indigo 3050 liquid EP press - 11”x17”, 7-

color, 2K sheets/hr. 4-color) (67 8.5”x11” ppm)

Gelato Strategy Council Meeting 24-26 May 2004

 Digital publishing is important
because…
printing is not going away
z Commercial printing revenue has
approximately doubled during each of
the past 10 years
Š 1983 – $92.8 B
Š 1993 – $176.6 B
Š 2003 – $330.5 B
z Paper has features that no other
media can offer
Š visual and tactile aesthetics
Š scalability in cost, distribution, and
size
Š flexibility and convenience

Gelato Strategy Council Meeting 24-26 May 2004

 Commercial printing generates a
wide range of products
z Books
z Periodicals
z Catalogs
z Advertising and solicitation
z Financial and legal
z Packaging
z Technical documents, instructions, and manuals

Gelato Strategy Council Meeting 24-26 May 2004

 Digital publishing is important
because…
it will transform commercial printing
z New short-run capabilities
Š print one copy or a thousand copies
Š print-on-demand
Š distribute-then-print
z Variable data printing
Š every copy can be different
Š pull content from databases
Š reformat each copy
z Workflow

Gelato Strategy Council Meeting 24-26 May 2004

 Stages 1 and 2 of the
commercial printing workflow

z Document creation
Š concept
Š content
Š design
Š layout
z Prepress preparation
Š impose
Š preflight
Š proof
Š generate plate

Gelato Strategy Council Meeting 24-26 May 2004

 Stages 3 and 4 of the commercial
printing workflow

z Press work
Š load paper
Š set up paper path
Š set up inks
Š monitor quality
z Finishing and distribution
Š cut
Š bind
Š ship

Gelato Strategy Council Meeting 24-26 May 2004

 Digital technologies have impacted
commercial printing, but…
z Print shops remain a cottage industry of small-
to-medium sized, highly capitalized businesses.
Š $5M to $10M sales
Š typically fewer than 20 employees
Š approximately 45K shops in U.S.
z Components of workflow have been separately
digitized. Paradigm hasn’t changed.
z Owners resist investment in digital technologies
due to fear of obsolescence and cost of
maintenance.
z Variable data printing is not well-understood or
widely used.
z Skilled labor is increasingly difficult to find.

Gelato Strategy Council Meeting 24-26 May 2004

 What does high performance
computing and the grid offer?

z Virtually unlimited computational resources when needed,

and only when needed.
z Uniform framework for locating and accessing resources.
z Global connectivity.

Gelato Strategy Council Meeting 24-26 May 2004

 Thin-client print shop concept
z Many print shop services and resources become virtual
and grid-based.
z The print shop itself may have only limited on-site
resources.
z Seamless integration between document designer, content
provider, printer, and customer.
z Fully supports “distribute-then-print” as opposed to
traditional “print-then-distribute” model.
z More complete support for variable data printing.

Gelato Strategy Council Meeting 24-26 May 2004

 What about high-performance
computing?
Four areas of opportunity
z Raster image processing
Š Text
Š Graphics/line art
Š Images
z Image rendering
z Artifact recognition
z Image processing for proofing
Š Emulate press on display or local printer
Š Show characteristics of target press
Š Hide limitations of proofing device

Gelato Strategy Council Meeting 24-26 May 2004

 Raster image processing (ripping)

z In order to print a document, we must transform from a

high-level page description language (postscript or PDF) or
document description language (PPML) to a raster image
suitable for the printer (hardware-ready bits)
z At 800 dpi, a single 8.5”x11” page contains 60 M pixels
z Printing at 67 ppm, we must process 66 M pixels/sec
z What do we have to do with each of those pixels?

Gelato Strategy Council Meeting 24-26 May 2004

 Imaging pipeline for digital printing

Image
processing
scale
crop
Generate sharpen
display list smooth
Document file Render objects gamut map
(page description on display list color convert
language, e.g. color correct
postscript) compress
decompress
.
.
.

Hardcopy Print Halftone

Gelato Strategy Council Meeting 24-26 May 2004

 If we can do it in a $500 desktop
printer, what is the big deal?
z Need higher speed
z Must deliver higher quality – no shortcuts
z Must handle wider range of content
z Opportunity to push the envelope for image quality

Gelato Strategy Council Meeting 24-26 May 2004

 Pushing the envelope:
Optimal image scaling by tree-
based resolution synthesis
T
Z X

W Estimate X from L
realization of Z

Source Scaled
Image Image

Gelato Strategy Council Meeting 24-26 May 2004

 Structure of estimator

CT (⋅):Z → {0 ,Κ , M − 1}
z

z Classify Aj , β j e0 , µ 0 e0t ( z − µ 0 ) < 0 ?

x̂
j = CT ( z ) yes no
e1, µ1 e2 , µ 2
xˆ = A j z + β j

j=0 j =1 e3 , µ 3 j=4
A0 , β 0 A1, β1 A4 , β 4

j=2 j=3
A2 , β 2 A3 , β 3

Gelato Strategy Council Meeting 24-26 May 2004

 4X scaling results

Photoshop Bicubic Interpolation Tree-Based Resolution Synthesis

Gelato Strategy Council Meeting 24-26 May 2004

 Pushing the envelope:
digital halftoning
z Point processes - screening
z Neighborhood processes - error diffusion
z Iterative processes - direct binary search (DBS)

Di

DBS screen Error diffusion DBS

Increasing complexity

Increasing quality
Gelato Strategy Council Meeting 24-26 May 2004
 Direct Binary Search
Human visual system filter kernel
~
p(x)
~
f(x)

f[n]
+ ~

Human visual system filter kernel ∑ e(x) arg min

g[n]
∫ ( e(x) ) dx
x
~ 2

~ -
p(x)
~
... g(x)

g0[n] g[n]

Gelato Strategy Council Meeting 24-26 May 2004

 The Search Heuristic
Toggle

Swap 1 Accept
pattern
Swap 2
with
lowest
error
Swap 3

Gelato Strategy Council Meeting 24-26 May 2004

 DBS Convergence: 0, 1, 2, 4, 6, and 8
Iterations

Gelato Strategy Council Meeting 24-26 May 2004

 A use case:
NAMI West Central Indiana
z Small, struggling, but forward-looking, “start-up” not-for-profit
organization in Lafayette, Indiana.
z Mission is to help those afflicted by serious mental illnesses and
their families, to educate the community about the nature of
serious mental illnesses, and to transform the mental health
care system.
z Mailing 2 pieces/month to 1200 addresses – database of 2500
contacts.
z Printing with an HP LaserJet 8550 (6 pages/minute 4-color).

Gelato Strategy Council Meeting 24-26 May 2004

 NAMI WCI monthly mailings

Gelato Strategy Council Meeting 24-26 May 2004

How can high performance computing
and the grid help NAMI WCI?
z Time and resources spent as an internal print shop could
be better spent helping consumers and their families.
Š more effective tools for document processing workflow
Š viable options for outsourcing of printing while allowing
involvement in creative process
z Better customization of mailings could more effectively
connect to community and increase level of engagement
and support.

Gelato Strategy Council Meeting 24-26 May 2004

 Thanks for your interest!

Gelato Strategy Council Meeting 24-26 May 2004