Abstract: A previous Interfaces (Edelman Finalist) paper [12] described the LDP Lean Document Production toolkit and work process based on operations research techniques that demonstrated significant improvements in the productivity of small to mid-sized print shops. In this paper, the extensions of the methodology to improve productivity of very large print shops are described. The printing industry is segmented into four quadrants based on resource utilization and job variability. Job variability in one of the industry segments is extremely high and is characterized by fat-tail distributions. This paper describes the extensions required to handle the applications of LDP Lean Document Production to this market segment.

1. International Journal of Performability Engineering Vol. 9, No. 1, January 2013, pp. 85-96.
© RAMS Consultants
Printed in India
_______________________________
*Corresponding author’s email: sudhendu.rai@xerox.com 85
Implementation of Lean Document Production in the
Printing Industry
SUDHENDU RAI*
Xerox Research Center Webster
MS 128-51E, 800 Phillips Road, Fairport, New York 14450, U.S.A.
(Received on January 01, 2012 and revised on September 21, 2012)
Abstract: A previous Interfaces (Edelman Finalist) paper [12] described the LDP Lean
Document Production toolkit and work process based on operations research techniques
that demonstrated significant improvements in the productivity of small to mid-sized print
shops. In this paper, the extensions of the methodology to improve productivity of very
large print shops are described. The printing industry is segmented into four quadrants
based on resource utilization and job variability. Job variability in one of the industry
segments is extremely high and is characterized by fat-tail distributions. This paper
describes the extensions required to handle the applications of LDP Lean Document
Production to this market segment.
Keywords: manufacturing: performance/productivity; production scheduling:
applications; statistics: data analysis; heavy-tail distributions
1. Introduction
It is well-known that the printing industry is large and fragmented. The North American
Industrial Classification System (NAICS) code for “printing and related support
activities” is 323. In 2005, the total value of print shipments corresponding to code 323
was $97.095 billion with an annual payroll of $24.893 billion [22]. The industry employed
642,300 employees with the payroll per employee of $38,753. An estimate of $100,000 in
annual sales per employee is remarkably accurate in determining a commercial printer’s
annual sales [21]. Changes over time in the numbers of small (1-9 employees), medium
(10-49 employees), and large (50+ employees) establishments provide a measure of
industry dynamics. In Figure 1, we show the number of print shops grouped by the
number of employees. The increase in the number of larger establishments and decline in
the number of small and medium shops reveals that business is moving from small and
medium sized shops to large shops.
Small Shops Medium Shops Large ShopsSmall Shops Medium Shops Large Shops
Figure 1: Commercial, Quick Printers by employee size, 2002 vs. 2004 vs.
2006. [21]

2. 86 Sudhendu Rai
The top business challenges faced by print and prepress firms are indicated in Figure 2.
The top three, economic conditions, competition, and pricing, all are symptoms of the
intense competitive dynamics of the industry. Since small and medium size shops are
diminishing in number and large shops are growing, the role of Operations Research in
the printing industry seems likely to increase. One of our major objectives in this paper is
to present an extension of Operations Research methodologies to handle the fat-tailed job
distributions characteristic of many of these increasingly dominant large shops.
Figure 2: Top business challenges faced by all print and prepress firms in the
fall, of 2006. [21]
2. Market Segmentation
Because of the great diversities in print shops, we developed a categorization scheme that
allows us to develop optimization tools and techniques for various segments of the print-
shop market. Based on the experience from early engagements, we developed the print-
shop market segmentation matrix shown in Figure 3. Print shops are characterized based
on the job variability (characterized by job mix, job size distribution and job inter-arrival
time distribution) and resource (equipment and manpower) utilization levels.
.
II. FM business,
commercial
printers, CRDs
IV . Large complex
operations
III . Document factories
or book printers
I. Mom and pop
shops
Utilization
II. FM business,
commercial
printers, CRDs
IV . Large complex
operations
III . Document factories
or
ook printers
I. Mom and pop
shops
Variability
Figure 3: A segmentation of print shops within the printing industry. FM
designates facilities management. CRD designates corporate reprographics
department.

3. Implementation of Lean Document Production in the Printing Industry 87
Segment I encompasses shops that operate individually to produce a few types of
products on as-needed basis (e.g., store-front print shops for convenience document
production). Segment II encompasses shops that are moderately sized (e.g., corporate
reprographics departments), produce several different types of job types (typically less
than forty), and are challenged with delivering high quality of service such as turnaround
time and print quality at competitive costs. Segment III encompasses shops that typically
specialize in a few different types of workflows to create products that are manufactured
in high-volumes (e.g., large book manufacturers). Typically these shops are found to
operate at higher levels of resource utilization than print shops found in segments I and II.
XMS operates shops in segments I, II and III.
Segment IV contains shops that manufacture a wide array of documents often within a
specific industry segment such as financial or healthcare, are large in size (e.g., over
$50M of annual revenue), and exhibit leveraged economies of scale in production
processes to achieve better utilization of resources than the shops in segments I and II.
LDP consulting engagements are offered to facilities in this segment. During the course
of the development of LDP Solutions print shops from each of the four segments were
engaged. Most effort was, however, focused on print shops in segments I, II and III.
Segment IV is the subject of ongoing research and development activity, the initial results
of which are described below.
Description of the results of implementations of the LDP techniques to quadrants I, II
and III of the market-segmentation matrix have been described in [12]. In this paper we
describe the extension of the LDP Lean Document Production techniques to the more
complex document factory consulting operations characteristic of Segment IV and of the
results obtained in the field from utilizing these extensions. .
3. The LDP (Lean Document Production) Methodology
The LDP methodology ([4],[6],[10],[12],[13],[14],[15],[16],[17],[20]) is used for
production management to make the document production process more efficient in terms
of high resource utilization, low waste, low Work In Process (WIP) and low inventory. It
utilizes the concept of autonomous cells [13] combined with novel job-splitting, routing
and event-driven scheduling, WIP management approaches using finite inter-process
buffers [7], and labor cross-training to improve the productivity of print shops. This
methodology has been described in some detail in a previous paper [12] to which we refer
the reader for details.
4. Improving Complex Document Factory Operations
Shops in quadrant IV are characterized both by higher levels of variability and/or product
variety and by higher levels of labor and equipment utilization (at least at bottlenecks)
when compared to shops in other quadrants. They are most likely to be large production
operations. These shops often employ industrial engineers who utilize manufacturing
engineering tools and methods to improve productivity. They are customers for
sophisticated optimization analyses produced by expert teams of operations research
consultants or industrial engineers. Providing optimization analyses for these shops is a
qualitatively different task than performing such analyses for the engagements described
above for quadrants I, II and III.
In Figure 4, we show a bar chart description of a shop that processes 961 different
product workflows. The process cycle efficiency (measured as the ratio of the value added
time to the total process cycle time) of such large shops is often quite low (often less than
10%). While many factors contribute to the low efficiency of these shops, high product

4. 88 Sudhendu Rai
variety can be an important one. Prior work [3] reports the potential of phase transitions in
scheduling efficiency whereby the scheduling efficiency of shops drops dramatically
when product workflow variety increases.
Job Count by unique workflow
0
20
40
60
80
100
120
140
160
1 81 161 241 321 401 481 561 641 721 801 881 961
Unique Job Type ID
NumberofJobsin
Category
Job Count
Figure 4: Print shop showing the job count associated with each workflow
type.
The coefficient of variation of the job-size distributions can be very high. Moreover, in
some cases the distribution may be fat-tailed. Other sources of operational complexity
arise from the sheer size of these shops. They can employ over 100 staff; have up to 100
pieces of equipment with sequence-dependent setup requirements; and process up to 1000
jobs per day with cycle time requirements that vary between a few hours to few days.
High variability as characterized by fat-tailed distributions in job sizes, job inter-arrival
rate and job mix has been observed in - large transaction and commercial print shops (e.g.,
those with annual revenues near or above $100 million). Although fat-tail phenomena
have been observed in Internet traffic [1] and computer workload on distributed
processors [9] this behavior has not previously been reported in the manufacturing
systems literature. In sections 7,8 and 9 we describe the extensions of LDP required to
apply it successfully to the large shops in quadrant IV that exhibit fat-tailed job
distributions. In the remainder of this section we describe the results of an initial
exploration of the use of these methods in one particular large shop.
The job size distribution of this particular shop exhibited fat-tail behavior. Its
equipments required up to six different types of setup depending on the sequence of jobs
processed on them. The shop had a very high WIP level, typically WIP in excess of 3
million units on a given day. The LDP proposed productivity improvement workflow
change was to separate small jobs from the large jobs and to dedicate autonomous cells to
handle each class separately. Detailed simulation models were constructed to determine
the structure of the autonomous cells and the corresponding scheduling policy. In Table 1
we show the results of implementing LDP on the production operations. This example
highlights the fact that application of LDP Solutions results in significant productivity
improvements for large, complex document production facilities as well as the smaller
and/or simpler shops characteristic of quadrants I, II and III in Figure 3.

5. Implementation of Lean Document Production in the Printing Industry 89
Table 1: Impact of implementation of LDP in a large production operation
ROI (one year) 178%
IRR- Internal Rate of Return 186%
Payback Period 11 months
Space Savings 15%
Product Travel Distance Reduction 75%
Throughput Time Savings 31%
Defects Per Million Reduction 55%
Productivity Improvement 12%
Sigma Metric Improvement 4.8 to 5.0
Yield Improvement 99.941 to 99.973
5. Shop and Scheduling Policy Design in the presence of Fat-tail Inputs
Large print shops are the first manufacturing systems reported to exhibit power-law
distribution characteristics, and in many instances – a fat tail. An example is given in
Figure 5.
Figure 5 : Fat-tailed job size distribution from a print shop
Let X be a random variable with cdf (cumulative density function) F(x) = P[X≤ x] and
complementary cdf (ccdf) Fc(x) = P[X>x]. We say here that a distribution F(x) is fat-
tailed if
20,~)( −

cxxFc (1)
In the limit of x->∞
−=
→x
c
xd
xdLogF
log
)(
lim
(2)
The theory of stable distributions [5],[19] has been used to characterize the behavior of
these distributions. A discussion of efficient methods for estimating the parameters and
properties of these distributions can be found in [11].
Fat-tailed distributions behave quite differently from distributions such as normal
distributions or exponential distributions that are most frequently used in characterizing
manufacturing system behavior. Because their tails decline relatively slowly, the

6. 90 Sudhendu Rai
probability of outlying observations is not negligible. Fat-tail distributions have infinite
variance indicating high variability in the underlying process that generates these
distributions. Because of these attributes, shop design in the presence of these
distributions differs from that of most shops as described below.
A simple method to detect the presence of fat-tail distribution is to plot ln(CCDF) and
ln(JobSize) as shown in Figure 6. As seen from equation (2) the negative value of the
slope for large job sizes will be equal to −. The slope of curve in Figure 6 towards the
end of the curve (i.e., for large job sizes) is -0.56 implying a fat-tail distribution.
Fat tailed distributions have many small jobs mixed with a few large jobs. Therefore,
even though most of the job-sizes are small, the major contribution to sample mean or
variance comes from the few large observations. In such distributions, the difference
between means and medians is usually very high. e.g., for the print shop data shown in
Figure 5, the mean = 17170 and median = 260.
Heavy-tailed distribution determination
-6
-5
-4
-3
-2
-1
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Ln(JobSize)
Ln(CCDF)
Heavy-tailed distribution
determination
Figure 6: Log(CCDF) and Log(JobSize) plot of job Size data from print shop
6. Calculation of Moments and Steady State
To analyze the behavior of the sample mean, we are concerned with the convergence
properties of sums of random variables. The normal starting point for such discussions
would be the Central Limit Theorem (CLT). Unfortunately, the CLT applies only to sums
of random variables with finite variance, and so does not apply in this case. In the place of
the CLT we instead have limit theorems for fat-tailed random variables first formulated by
L´evy [5],[19].
To introduce these results we need to define the notation A
d
→ B which means that
the random variable A converges in the distribution to B (roughly, has distribution B for
large n). Then the usual CLT can be stated as: for Xi i.i.d and drawn from some
distribution F with mean  and variance 
∞ define
=
=
n
i
in X
n
A
1
1
(3)
)(2/1
−= −
nn AnZ (4)
then
),0( 2
→
d
nZ (5)

7. Implementation of Lean Document Production in the Printing Industry 91
in which N(0,2
) is a Normal distribution.
However, if Xi are i.i.d. and drawn from some distribution F that is fat-tailed with tail
index 1<  <2, then if we define
)(/11

−= −
nn AnZ (6)
we find that
SZ
d
n → (7)
where S is an −Stable distribution [19].
The consequence of equation (6) is that as  gets smaller, the convergence of the
sample mean to the population mean becomes very slow – much slower than in the case of
distributions with finite variance. In Figure 7 we show the sample mean calculations
performed on samples drawn from two distributions – one being normal and another that
is fat-tailed. It shows how the mean of the sample drawn from normal distribution
converges at 100 data-points but the mean of samples drawn from a fat-tail job size
distribution of a large print shop has not converged even when the sample size is 2000.
Simulations performed with fat-tailed inputs [2] converge slowly to steady state and show
high variability at steady state. The practical consequence is that it becomes difficult for
such print shops to predict performance within tight bounds.
7. Scheduling
A prior paper [8] discusses various scheduling policies when task-size distribution is fat-
tailed. They consider four task-assignment policies for such distributed server systems:
Round Robin; Random; Size-Based, in which all tasks within a given size range are
assigned to a given host; and Dynamic-Least-Work-Remaining, in which a task is
assigned to a host with least outstanding work. They reported that when the task-sizes are
not highly variable, the Dynamic-Least-Work-Remaining policy is preferable. However,
when the task-size distribution is fat-tailed and exhibits high variability, then the Size-
Based policy is the best choice.
Figure 7: Mean calculations on samples drawn from a normal distribution
and a fat-tail job-size distribution.
Unlike the single stage computer systems discussed in [8], the large transaction print
shops where fat-tailed behavior was observed have multiple production stages and also
sequence dependent setups. Print jobs arrive in the print shop and are sent to the printer
queues for printing. Subsequently, they are moved to inserting equipment for folding and

8. 92 Sudhendu Rai
insertion into envelopes. A setup in the form of roll-changeover is incurred each time the
underlying pre-printed form changes. Similarly, a new setup is needed when requirements
on the inserting machines change. The combined effect of sequence-dependent setups and
fat-tail job size distribution creates difficulties in achieving high throughput in these
shops.
In Figure 8 we show the scheduling architecture that we studied for creating cells and
scheduling jobs in large print shops with fat-tailed inputs. In step 1, the incoming job
stream is separated into two sets using a job-size threshold parameter – one that contains
small jobs and the other that are jobs above the threshold. Autonomous cells are designed
to process the small jobs in their entirety. The small jobs are further sorted and grouped
based on common attributes to reduce setup time on machines. The larger jobs are split
into two sets – one that requires less frequent setup and the other that requires more
frequent setups. Multiple strategies are used to perform this partitioning. In one strategy,
the number of unique setup requirements is determined separately for multiple production
steps (for e.g. printer setup and insertion setup) for each job. Each job is assigned a
number that is the maximum of the setup numbers for each processing step. A threshold
value is picked. Jobs that have assigned number that is less than the threshold are put in
the low-setup pool, and the rest are put in a high-setup pool. The threshold value is
iterated upon using multiple simulation studies to determine the value that optimizes the
print shops performance objective (e.g., number of late jobs, resource utilization,
makespan). Autonomous cells are then designed for each job stream. The key motivation
behind this strategy is to separate jobs that require high setups from the ones that require
low setup and to design different types of autonomous production cells to handle the two
types.
Job arrival
Job classifier by size
Big job pool classifier Small job pool scheduler
Threshold for job stream splitting based on attributes
thatdetermine setup time
High setup job
router
Low setup job
router
Cell1
Queue
sequencing
Batch Splitting, Sorting,
Machine assignment
Cell1
Queue
sequencing
Batch Splitting, Sorting
Machine assignment
Cell1
Queue
sequencing
Batch Splitting, Sorting,
Machine assignment
Cell1
Queue
sequencing
Batch Splitting, Sorting
Machine assignment
Job size threshold parameter
Cell1
Queue
sequencing
Batch Splitting, Sorting,
Machine assignment
Cell1
Queue
sequencing
Batch Splitting, Sorting
Machine assignment
Figure 8: A scheduling architecture for large print shops with fat-tailed job
size inputs
8. Heuristic Solution
High-variability in job size distribution can lead to high queue waiting times. In Figure 9
we show a G/G/m queue where job arrivals are routed to a set of identical printers. G/G/m
queue refers to a system of m parallel machines with general (denoted by G) inter-arrival

9. Implementation of Lean Document Production in the Printing Industry 93
and processing time distributions. The waiting time in the queue is directly proportional to
the sum of squares of coefficient of variation of the inter-arrival time distribution and job
size distribution respectively.
Splitting jobs into smaller batches has the effect of lowering the coefficient of variation
of the job size distribution but increasing the coefficient of variation of the inter-arrival
time distribution. If the effect of batch-splitting leads to an overall reduction in the value
of (ca
2
+ce
2
)/2, there will be a reduction in the queue waiting times at each printer.
Figure 9: Queue waiting time in a G/G/m queue [23] with job arrivals routed
to a set of identical printers
9. An Example
In Figure 10 we show the inter-arrival time distribution and the job size distribution from
a print shop that has high variability in job size distribution. The coefficient of variation
(CV) of the inter-arrival time distribution is 2.7 and the coefficient of variation (CV) of
the job-size distribution is 5.6.
6000005000004000003000002000001000000
Median
Mean
6000500040003000200010000
A nderson-Darling Normality Test
V ariance 665519107
Skewness 16.653
Kurtosis 333.111
N 1147
Minimum 0
A -Squared
1st Q uartile 0
Median 220
3rd Q uartile 2210
Maximum 614640
95% C onfidence Interv al for Mean
3088
314.09
6077
95% C onfidence Interv al for Median
160 325
95% C onfidence Interv al for StDev
24783 26899
P-V alue < 0.005
Mean 4582
StDev 25798
95% Confidence Intervals
Summary for B&W Production PPS-300
2824201612840
Median
Mean
0.80.60.40.20.0
A nderson-Darling Normality Test
V ariance 3.3198
Skewness 7.4042
Kurtosis 75.1265
N 1134
Minimum 0.0000
A -Squared
1st Q uartile 0.0000
Median 0.0800
3rd Q uartile 0.7300
Maximum 28.2500
95% C onfidence Interv al for Mean
0.5680
204.22
0.7804
95% C onfidence Interv al for Median
0.0600 0.1100
95% C onfidence Interv al for StDev
1.7500 1.9003
P-V alue < 0.005
Mean 0.6742
StDev 1.8220
95% Confidence Intervals
Summary for Inter-Arrival Time
CV = 5.6 CV = 2.7
6000005000004000003000002000001000000
Median
Mean
6000500040003000200010000
A nderson-Darling Normality Test
V ariance 665519107
Skewness 16.653
Kurtosis 333.111
N 1147
Minimum 0
A -Squared
1st Q uartile 0
Median 220
3rd Q uartile 2210
Maximum 614640
95% C onfidence Interv al for Mean
3088
314.09
6077
95% C onfidence Interv al for Median
160 325
95% C onfidence Interv al for StDev
24783 26899
P-V alue < 0.005
Mean 4582
StDev 25798
95% Confidence Intervals
Summary for B&W Production PPS-300
2824201612840
Median
Mean
0.80.60.40.20.0
A nderson-Darling Normality Test
V ariance 3.3198
Skewness 7.4042
Kurtosis 75.1265
N 1134
Minimum 0.0000
A -Squared
1st Q uartile 0.0000
Median 0.0800
3rd Q uartile 0.7300
Maximum 28.2500
95% C onfidence Interv al for Mean
0.5680
204.22
0.7804
95% C onfidence Interv al for Median
0.0600 0.1100
95% C onfidence Interv al for StDev
1.7500 1.9003
P-V alue < 0.005
Mean 0.6742
StDev 1.8220
95% Confidence Intervals
Summary for Inter-Arrival Time
CV = 5.6 CV = 2.7
Figure 10: Job size distribution and inter-arrival time distribution of black-and-white jobs
from a print shop. The coefficient of variation is indicated by CV.
Table 2 indicates the effect of batch splitting on shop performance. While the CV of the
job size distribution gets reduced significantly, the CV of the split-job distribution

10. 94 Sudhendu Rai
experiences only a moderate increase. The net result is that the value of the term
(ca
2
+ce
2
)/2 is reduced from 19.3 to 6.2 (i.e. a reduction by a factor of 3.1).
This example illustrates the fact that the result of splitting very large jobs (in the fat-tail
distribution) into smaller batches prior to routing these to autonomous cells has the effect
of reducing queue waiting times. Further discussion of a batch-splitting heuristics in the
presence of high-variability job size inputs can be found in [18]. In addition to performing
batch-splitting prior to routing jobs to cells, the batches that are processed within a cell
can be further split according to policies discussed in [12] to minimize work-in-progress
and improve flow and cycle time.
Table 2: Changes in job size and inter-arrival distributions after batch
splitting.
2.25.6CV
969831017Stdev
44036782Mean
After
Splitting
OriginalJob Size
2.25.6CV
969831017Stdev
44036782Mean
After
Splitting
OriginalJob Size
2.762.70CV
1.831.87Stdev
0.660.69Mean
After
Splitting
OriginalInter-Arrival
Time
2.762.70CV
1.831.87Stdev
0.660.69Mean
After
Splitting
OriginalInter-Arrival
Time
Efficient sequencing policies to assign production priority to jobs queued up at the
various cells are implemented to optimize desired production requirements like
minimizing the percentage of late jobs or improving utilization. The interaction of these
variables and their resulting impact on cell design, scheduling policies and performance
can be quite complex. Extensive simulation models were built that allow iteration and
optimization of the parameters of the architecture shown in Figure 8.
The solution described by the architecture diagram in Figure 8 is a function of the job
size, setup requirements, and related characteristics of the incoming job stream. As the
customers’ job mix changes, the cells may require re-adjustment and optimization. While
the proposed architecture delivers superior performance to a functionally organized print
shop, it needs ongoing maintenance and optimization as the job mix changes over time.
10. Summary
This paper describes a four quadrant framework (Figure 3) for characterizing print shops
in the printing industry based on the workflow complexity and utilization of resources.
While an earlier paper [12] described the application of a productivity improvement
solution namely, LDP Lean Document Production to three of the four quadrants
(Quadrant I, II and III), the focus of this paper is on large shops (Quadrant IV) that exhibit
high levels of workflow complexity and typically run at higher levels of utilization than
shops in other quadrants. The formalism of heavy tail distributions is used to characterize
the extreme levels of variability observed in job size. Methods of quantifying and
characterizing the variability are presented. A production architecture (Figure 8) is
described along with scheduling policy heuristics that improve the productivity of such
shops. A simulation-based optimization framework is utilized to evaluate and optimize the
various parameters associated with the architecture and scheduling policies. Data analytics
from a large print shop where this heavy tailed job size distribution was observed is
described. The implementation of the LDP Lean Document Production solution in a large

11. Implementation of Lean Document Production in the Printing Industry 95
transaction shop is presented (Table 1) and is shown to deliver substantial productivity
benefits.
References
[1]. Crovella, M.E., Taqqu, M.S., and Bestavros, A. 1998. Heavy-tailed probability distributions
in the world wide web. A Practical Guide to Heavy Tails, Chapter 1, Pages 1-23. Chapman
& Hall. New York.
[2]. Crovella, M.E. and Lipsky, L. 1997. Long Lasting transient conditions in simulations with
heavy-tail workloads. Proceedings of the 1997 Winter Simulation Conference.
[3]. Donati, A.V., Darley, V., and Ramachandran, B. 2004. An Ant-Bidding Algorithm for
Generalized Flow Shop Scheduling Problem: Optimization and Phase Transitions. (Working
paper) www.santafe.edu/~vince/.
[4]. Duke, C. B., Jackson, W. B., and Rai, S. 1999. Interactive distributed communication
method and system for bidding on, scheduling, routing and executing a document processing
job, US patent 6,573,910, filed in Nov. 1999, granted in June 2003.
[5]. Feller, W.1971. An Introduction to Probability Theory and Its Applications. Volume II.
Second edition. John Wiley and Sons.
[6]. Gartstein, Y. 2001. Systems and methods for optimizing a production facility, US patent
6,633,790, filed in Jan. 2001, granted in Oct. 2003.
[7]. Gershwin, S.B. 1994. Manufacturing Systems Engineering. Prentice-Hall, Englewood Cliffs,
NJ. ISBN: 0-135-60608X
[8]. Harchol-Balter, M., Crovella, M.E., Murta, C.D. 2004. On Choosing a Task Assignment
policy for a distributed server system. Lecture Notes in Computer Science. Springer
Berlin/Heidelberg, ISBN 0302-9743.
[9]. Harchol-Balter, M. and Downey, A. 1996. Exploiting process lifetime distributions for
dynamic load balancing. SIGMETRICS ’96, pages 13-24..
[10]. Jackson, W. B. and Rai, S. 2000. System and method for converting print jobs stored in
print shop job description language files into print shop workflow, US patent 7,064,848,
filed in Dec. 2000, granted in June 2006.
[11]. Nolan, J.P. 1998. Multivariate stable distributions: approximation, estimation, simulation
and identification. In R.J. Adler, R.E. Feldman, and M.S. Taqqu (Eds.), A Practical Guide to
Heavy Tails, pp. 509-526. Boston: Birkhauser.
[12]. Rai, S., Duke, C. B., Lowe, V., Quan-Trotter, C., and Scheermesser, T. 2009. LDP Lean
Document Production®
– O.R. enhanced productivity improvements for the printing
industry. Interfaces. Vol. 39, No. 1, January–February 2009, pp. 69–90.
[13]. Rai, S., Godambe, A., Duke, C. B., and Williams, G. 2000. Print shop resource optimization
via the use of autonomous cells, US patent 7,079,266, filed in Nov. 2000, granted in July
2006.
[14]. Rai, S., Gurunatha, T. "Integrating LDP Lean Document Production solution within the
DMEDI methodology". International Journal of Performability Engineering. Vol. 6. Number
6. November, 2010.
[15]. Rai, S., and Maitra, S. C. 2001 Method for determining optimal batch sizes for processing
print jobs in a printing environment, US patent 6,805,502, filed in Jan. 2001, granted in Oct.
2004.
[16]. Rai, S. 2005. System and method for dynamically reconfiguring one or more autonomous
cells in a print shop environment. Patent publication 20070091355, Filed August, 2005
Published April, 2007.
[17]. Rai, S. and Viassolo, D. 2001. Production server architecture and methods for automated
control of production document management, US patent 7,051,328, filed in Jan. 2001,
granted in June 2006.
[18]. Rai, S. and Zhang, Z. 2006. Print job management system. US patent 7,755, Filed March,
2006, granted in July 2010.
[19]. Samorodnistsky, G. and Taqqu, M. 1994. Stable Non-Gaussian Random Processes.
Stochastic Modeling. Chapman and Hall. New York.

12. 96 Sudhendu Rai
[20]. Squires, D and Rai S. 2000. Production server for automated control of production document
management, US patent 7,065,567, filed in Nov. 2000, granted in June 2006.
[21]. The Industry Measure, 2007. Printing Market Demographic Profiles. The Industry Measure
updated snapshots of the print and prepress markets. January 2007.
http://www.theindustrymeasure.com
[22]. U.S. Census Bureau. 2008. The Statistical Abstract, 2008 Edition. The National Data Book.
http://www.census.gov/compendia/statab/2008edition.html
[23]. Whitt,W. 1983. The Queuing Network Analyzer. Bell System Technology Journal 62(9):
2779-2815.
Sudhendu Rai is a Xerox Fellow, Program Manager and a certified Lean Six Sigma
Black Belt at the Xerox Research Center in Webster, N.Y. He received his Ph.D. from
MIT in 1993, M.S. from Caltech in 1989, and B.Tech. from IIT, Kanpur (India) in 1988 –
all in Mechanical Engineering. He is the lead inventor of the LDP Lean Document
Production®
Solution that was selected as a finalist in the Franz Edelman (worldwide)
competition sponsored by INFORMS. This work has also been selected as one of the three
finalists for the Service Innovation Practice Award sponsored by SRII (Service Research
and Innovation Institute). LDP also won the first place at the IIE Lean Best Practices
Award in 2011. Starting in 1998, he led a team that developed the algorithms, software
toolkit to support the initial offering and a training curriculum to train Xerox Global
Services consultants. He has personally led and implemented process improvement
initiatives in dozens of small and large print shops spanning multiple industry segments.
He holds 37 patents (with 40 additional pending) and has published more than 25
technical papers in conference proceedings and technical journals. He is a senior member
of IEEE. He is a recipient of the Xerox Excellence in Science and Technology Award and
was selected as a finalist for the Rochester Engineer of the Year award in 2007.