1 logistics & supply chain management strategies peerapol sittivijan...

1

Logistics & Supply Chain Management Strategies

Peerapol Sittivijan

E-mail:[email protected]

August 19, 2009

Guest Lecture at KMUTNB

2

OutlineOutline

• What is supply chain & why it is important?

• The Value of Information: Bullwhip Effect

• Risk Pooling Strategy

• Push, Pull, Push-Pull Strategy

• Distribution Strategy

• Revenue Management

3

Introduction

©Copyright 2003 D. Simchi-Levi

Manufacturing Plants

RegionalWarehouses(DistributionCenters)

Field Wrhs(Dist. Centers)

RetailStores

Customers

A Typical Supply ChainA Typical Supply Chain

Suppliers


Example 1: Wal-Mart

Customers Request:Buying detergent,clothes, TV, …...

Wal-Mart Stores

Wal-Mart or third-partydistributioncenters

Procter & Gamble

Plastic ProducerFabric Producer

Da-Fa Clothing, Inc. (China)

SONY Factory (Malaysia)

Electronics Components Producer

Chemical Producer

Zipper Producer

Thread Producer

Plastic Producer

Supply Chain ExamplesSupply Chain Examples


Suppliers IC Mfg

PC Board

Subassembly

Suppliers

FAT

US DCs

Europe DCs

Far East DCs

Suppliers

Suppliers

Retailer

Retailer

Retailer

Example 2: Hewlett & Packard (HP)

FAT = Final assembly & testIC Mfg = Integrated circuit manufacturingPC Board = Printed circuit board

Consumers

Consumers

Consumers

Supply Chain Examples (Cont’d)Supply Chain Examples (Cont’d)


Example 3: Dell

Customers order computers on-line

Dell Assembly Plants

Monitors by SONY (Mexico)

Keyboards by Acer (Taiwan)

CPU by Intel (USA)

Other components

Supply Chain Examples (Cont’d)Supply Chain Examples (Cont’d)


• Definition:

Supply Chain Management is primarily concerned with the efficient integration of suppliers, factories, warehouses and stores so that merchandise is produced and distributed in the right quantities, to the right locations and at the right time, and so as to minimize total system cost subject to satisfying service requirements.

• Notice:– Who is involved– Cost and Service Level– It is all about integration

Supply Chain Management


Processes Involved in SCMProcesses Involved in SCM

Suppliers Manufacturers Distributors Retailers Consumers

Customer order1. Order arrival (retail stores call/mail order center, website)2. Order entry3. Order fulfillment4. Order receiving

Replenishment1. Retail order trigger (inventory low)2. Retail order entry3. Retail order fulfillment4. Retail order receiving

Manufacturing1. Order arrival from Distr.2. Production scheduling3. Manufacturing & shipping4. Order receiving by Distr.

Procurement1. Order from Mfg2. Supplier production scheduling3. Component manufacturing/shipping4. Order receiving by Mfg


TYPICAL DECISIONS

Strategic

Tactical

TYPETIME FRAME

•Supply chain strategies (Sell direct or through retailers? Outsource or in-house? Focus on cost or customer service?)•Supply chain network design (How many plants? Location and capacities of plants and warehouses?)•Product mix at each plant

years

•Workforce & Production planning •Inventory policies (safety stock level)•Which locations supply which markets•Transportation strategies

3 mo.- 1year

Operational•Production scheduling •Distribution scheduling and routing•Place inventory replenishment orders•Lead time quotations

daily

Supply Chain DecisionsSupply Chain Decisions


Conflicting local objectives across SC

Manufacturers Distributors Retailers Consumers

Convenience

Short lead time

Large variety of products

Few stores

Low inventory

Little variety

Close to DCs

Low inventory

Few DCs

Large shipments

Large production batches


Objective of SCMObjective of SCM

• SCM is concerned with the efficient management of a supply chain so as to maximize supply chain profitability across the entire supply chain

Supply chain profitability = total revenue - total cost

• Manage globally, not locally

• Consider both cost and customer service (Optimal tradeoff between cost and customer service)

(Companies do not compete, their supply chains do)


Procurement Planning

ManufacturingPlanning

DistributionPlanning

DemandPlanning

Sequential Optimization

Supply Contracts/Collaboration/Information Systems and DSS

Procurement Planning

ManufacturingPlanning

DistributionPlanning

DemandPlanning

Global Optimization

Sequential Optimization vs. Global Optimization


Success for SCM in PracticesSuccess for SCM in Practices

• P&G developed close relationship with their suppliers and jointly created business plans to eliminate wasteful practices across the entire supply chain. As a result, it could save $85 million in 18 months.

• Wal-Mart is the largest and highest-profit retailer in the world by initiating the cross-docking strategy. Goods are continuously delivered to warehouses without ever sitting in inventory.

© Zhi-Long Chen 15

The Value of Information:Bullwhip Effect


Information AgeInformation Age

• Internet, Wi-Fi, XML, Web service and Data warehouse

• Provide abundance of available information

• Opportunity to effectively design and manage integrated supply chain

• “In modern supply chain, information replace inventory”


The Bullwhip EffectThe Bullwhip Effect

• End-customers’ demand does not vary much, however, inventory and back-order levels considerably fluctuate

• the wholesaler receives orders from the retailer and place orders to the distributor

• Without end-customers’ demand data, the wholesaler use the order from the retailer to perform the forecast and make orders

External Demand

Retailer

Order lead time

Wholesaler

Order lead time

Distributor

Order lead time

Manufacturer

Production lead time

Delivery lead time

Delivery lead time

Delivery lead time


The Bullwhip Effect (Cont’d)The Bullwhip Effect (Cont’d)

• Since variability in orders by the retailer is much higher than the end-customers’ demand, wholesaler is forced to carry more safety to maintain the same service level as retailer

• This results in higher inventory levels and costs at higher stages in a supply chain

The Bullwhip Effect (Cont’d)The Bullwhip Effect (Cont’d)


Ord

er

Siz

e

Time

CustomerDemand

CustomerDemand

Retailer OrdersRetailer OrdersDistributor OrdersDistributor Orders

Production PlanProduction Plan

Example 1: P&G DiapersExample 1: P&G Diapers


Lee, H, P. Padmanabhan and S. Wang (1997), Sloan Management Review


Factors Affecting VariabilityFactors Affecting Variability

• Demand Forecasting: Inventory policies such as s,S uses the forecast of mean and S.D. of demand to make orders so higher S.D. of demand, higher variability of order quantities

• Lead Time: With a longer lead time, a small change in estimate of demand variability is magnified in significant change in

Order-up-to-level = Q + AVG · L + z · STD · L

Average demand over lead time

Safety stock


Factors Affecting Variability (Cont’d)Factors Affecting Variability (Cont’d)

• Batch ordering: A large order followed by several periods of no orders and followed by another period of a large order

• Price fluctuation: Retailers tends to stock up inventories when prices are lower. This can lead to batch ordering

• Inflated orders: Retailers tends to stock up inventories when they suspect that products will be in short supply. This can also lead to distortion of demand estimation


Methods to Cope with Bullwhip EffectMethods to Cope with Bullwhip Effect

• Reducing uncertainty: By centralizing the demand information throughout the SC, each stage of the SC can perceive the complete information on actual end-customers’ demand

• Reducing variability: Reduce the variability of end-customers’ demand, for example, everyday low pricing

• Lead time reduction: Cross-docking to reduce delivery lead times and EDI to reduce information lead time


Centralizing Demand InformationCentralizing Demand Information

• Centralized demand information: each stage of a supply chain receives the customers’ demand information at the POS and uses it to plan their inventory and production plans

2

k

1i

2i

k

1iik

p

)L2(

p

L2 1

Var(D)

)Var(Q

k

1i2

2ii

k

]p

2L

p

2L [1

Var(D)

)Var(Q

Li = Lead time from stage i to i+1

p = No. of customers’ demand observation


Retailer-Supplier PartnershipsRetailer-Supplier Partnerships

• Quick response: suppliers receive POS data from retailers and use it to prepare production and inventory activities, however, the retailers still prepare individual orders

• Continuous replenishment: similar to quick response but suppliers use the data to prepare shipments at predetermined levels of inventory

• Vendor-managed inventory (VMI): the suppliers decide on the appropriate inventory policies

Efficiency Trust

© Zhi-Long Chen 26

Risk Pooling

© Zhi-Long Chen27

Risk PoolingRisk Pooling

• Consider two systems:

Warehouse 1

Warehouse 2

Market 1

Market 2

Supplier

Decentralized System:Two warehouses,each serving one customer

WarehouseMarket 1

Market 2

SupplierCentralized System:One warehouse,serving both customers

Questions: Q1: For the same service level, which system will require more inventory?Q2: For the same total inventory level, which system will have better service?

© Zhi-Long Chen28

• Compare the two systems:– one product– maintain 97% service level– $60 fixed order cost– $0.27 weekly holding cost– 1 week lead time– historical data on demand available (see table on next slide),

assume these data correctly represent demand distributions

Risk PoolingRisk Pooling

© Zhi-Long Chen 29

Risk Pooling ExampleRisk Pooling Example

Week 1 2 3 4 5 6 7 8

Market 1 33 45 37 38 55 30 18 58

Market 2 46 35 41 40 26 48 18 55

Total 79 80 78 78 81 78 36 113

• Historical demand data

© Zhi-Long Chen 30


AVG STD

Market 1 39.3 13.2

Market 2 38.6 12.0

Total 77.9 20.7

• Summary of historical data

Decentralized WH1

Decentralized WH2

Centralized WH

© Zhi-Long Chen 31

Given n observations of a random variable, X1, X2, ..., Xn,find mean, variance, and coefficient of variation

M ean n

XX

n

ii

1

V ariance: 1

)(1

2

2

n

XXn

ii

Standard deviat io n: 1

)(1

2

n

XXn

ii

C o effic ient o f var iat io n: X

CV

Excel: Function AVERAGE()

Excel: Function STDEV()

Quick Review of StatisticsQuick Review of Statistics

© Zhi-Long Chen32

(s, S) Policy(s, S) PolicyIn

vent

ory

Lev

el

S

s

0

LeadTimeLeadTime

Inventory Position

Q*

• Use EOQ model to determine optimal order quantity

– Q* =

– Set S = Q* + s

2 K·AVG

hWhereK = setup (ordering) costAVG = mean demand rateh = unit holding cost per unit time

SS

© Zhi-Long Chen 33

More specifically…. More specifically….

• Model with uncertain demand, constant lead timeL = constant lead timeAVG = mean demand rateSTD = stdev of demand rateAssume: Demand rate follows Normal(AVG, STD2)

Reorder point: s = AVG · L + z · STD · L

Average demand over lead time

Safety stock

Answer: Demand over lead time follows Normal(AVG·L, STD2 ·L)

Question: what is the distribution of demand over lead time?

© Zhi-Long Chen 34


AVG STD SS s Q SAverage

Inventory

Warehouse 1 39.3 13.2 25.08 65 132 197 91

Warehouse 2 38.6 12.0 22.8 62 131 193 88

CentralizedWarehouse

77.9 20.7 39.35 118 186 304 132

• Optimal inventory policies

Decentralized system: total SS = 47.88

total avg. invent. = 179

LSS = z ·STD ·s = AVG·L + SSQ = sqrt(2K·AVG/h)

Safety StockReorder PointOrder QuantityOrder-up-to-level S = s + Q

Average Inventory SS + Q/2

© Zhi-Long Chen35

Risk Pooling: Important Risk Pooling: Important ObservationsObservations

• Centralizing inventory control reduces both safety stock and average inventory level for the same service level.

(This phenomenon is called risk pooling)

• Root Cause: Demand Variability (i.e. STD) Variability of aggregated demand is lower than total variability of individual demands

SS = (z)(STD)(sqrt(L))Avg inventory = SS + Q/2

SS = (z)(STD)(sqrt(L))Avg inventory = SS + Q/2

Everything else being equal,a system with a lower demand variability requires a lower SS & lower avg inventory

Everything else being equal,a system with a lower demand variability requires a lower SS & lower avg inventory

Question: Why variability of aggregated demand is lower than total variability of individual demands ?

© Zhi-Long Chen 36

WarehouseMarket 1

Market 2

d1+d2: (, 2)

Calculating demand variability of Calculating demand variability of centralized systemcentralized system

Warehouse 1

Warehouse 2

Market 1

Market 2

d1: (1, 12)

d2: (2, 22)

2 = 1

2 + 22 + 212,

where -1 12

= 12 + 2

2 + 212,

where -1 1

: correlation coefficient of d1, d2

1+ 2 1+ 2

Conclusions: 1. Stdev of aggregated demand is less than the sum of stdev of individual demands2. If demands are independent or negatively correlated, the std of aggregated demand is much less

Conclusions: 1. Stdev of aggregated demand is less than the sum of stdev of individual demands2. If demands are independent or negatively correlated, the std of aggregated demand is much less

1. If d1, d2 positively correlated, > 02. If d1, d2 negatively correlated, < 0

= 1 + 2

= ??

1+2

10-1

22

21

P.C.N.C. Ind.

© Zhi-Long Chen 37

More ObservationsMore Observations

• In general, total safety stock & average inventory both increase with the number of stocking locations

Total SS (Avg. Inv.)

# of stocking locations

© Zhi-Long Chen 38

DecentralizedCentralized

Inbound transportation cost (from factories to warehouses)

Facility/Labor cost

Outbound transportation cost (from warehouses to retailers)

Inventory cost

Responsiveness to customers

Lower

Lower

Higher

Lower

Lower

Centralized vs. DecentralizedCentralized vs. Decentralized

© Zhi-Long Chen39

Critical Points about Risk PoolingCritical Points about Risk Pooling

• Centralizing inventory reduces both safety stock and average inventory in the system.

• The benefits from risk pooling depend on the behavior of demand from one market relative to the demand form another (demand correlations).

• The higher the coefficient of variation, the greater the benefit obtained from centralized system.

© Zhi-Long Chen 40

Push, Pull, Push-Pull Strategy


Push StrategyPush Strategy

• Production and distribution decision based on long term forecasts

• Take long time to react to the changing marketplace

• Problems

- Inability to meet changing demand patterns

- Product obsolescence

- Excessive inventories for large safety stock

- Low service levels


Pull StrategyPull Strategy

• Production and distribution based on demand driven by fast information flow to transfer demand data

• Coordinate with true customer demand rather than forecast demand

Advantages

- Reduced lead times (better anticipation and coordination)

- Decreased inventory levels at retailers and manufacturers

- Decreased system variability

- Better response to changing markets

disadvantages

- Harder to leverage economies of scale (manufacturing and transportation)

- Doesn’t work in the case that lead times are too long

© David Simchi-Levi

Matching Supply Chain Strategies Matching Supply Chain Strategies with Productswith Products

Pull Push

Pull

Push

I

Computer

II

Furniture, Automobile

IV

Books, CDs

III

Grocery

Demand uncertainty

(C.V.)

Delivery costUnit price

L H

H

L

Economies of Scale


Selecting the Best SC StrategySelecting the Best SC Strategy

• Higher demand uncertainty suggests pull• High importance of economies of scale suggests

push• High uncertainty/ EOS not important such as the

computer industry implies pull• Low uncertainty/ EOS important such as groceries

implies push– Demand is stable– Transportation cost reduction is critical– Pull would not be appropriate here.


Selecting the Best SC StrategySelecting the Best SC Strategy

• Low uncertainty but low value of economies of scale (high volume books and cd’s)– Either push strategies or push/pull strategies might be most

appropriate

• High uncertainty and high value of economies of scale– For example, the furniture or automobile industry– How can production be pull but delivery push?– Is this a “pull-push” system?


Push-Pull StrategyPush-Pull Strategy

• Pull strategy normally leads to a reduction in lead times, inventory level and system costs and better resource utilization

• Unfortunately, it is impractical to implement a pull-based system throughout the entire supply chain

• Due to too long lead times, necessary of economies of scale in production and transportation

• Therefore, a combination of push and pull strategies is preferred

End customerRaw materials Components Assembly

Push strategy Pull strategy

Dell’s push-pull strategy

© David Simchi-Levi47

Locating the Push-Pull BoundaryLocating the Push-Pull Boundary

• The push section:– Uncertainty is relatively low– Economies of scale important– Long lead times– Complex supply chain structures

• Thus:– Management based on forecasts is appropriate– Focus is on cost minimization– Achieved by effective resource utilization – supply chain

optimization


•The pull section:–High uncertainty

–Simple supply chain structure

–Short lead times

•Thus–Reacting to realized demand is important

–Focus on service level

–Flexible and responsive approaches



Characteristics and SkillsCharacteristics and Skills

RawMaterial Customers

PullPush

Low Uncertainty

Long Lead Times

Cost Minimization

Resource Allocation

High Uncertainty

Short Cycle Times

Service Level

Responsiveness



• The push section requires:– Supply chain planning– Long term strategies

• The pull section requires:– Order fulfillment processes– Customer relationship management

• Buffer inventory at the boundaries:– The output of the tactical planning process– The input to the order fulfillment process.

© Zhi-Long Chen 52

Distribution Strategy


Types of Distribution StrategiesTypes of Distribution Strategies

• Direct Shipping

• Shipping via Warehouses

• Shipping via Cross Docks

• Third Party Logistics (3PL)


Direct Shipping (Cont’d)Direct Shipping (Cont’d)

Advantages:

• Avoids the expense of operating a distribution center

• Lead times are reduced

Disadvantages:

• Risk pooling strategy cannot be applied in this case

• Transportation costs of manufacturing and retailer increase because inventory must be sent in smaller volumes (less than truck load)


Shipping Via WarehousesShipping Via Warehouses

Advantages:• Better service from regional locations• Transportation economies• Mixing functions• Risk pooling over the manufacturing or procurement

lead time• Differentiated stocking and service policies


Shipping Via Warehouses (Cont’d)Shipping Via Warehouses (Cont’d)

Design and planning issues:• Number of echelons• Number and location of distribution centers • Stock location: what items to stock at each DC• Replenishment policies – inventory & transportation;

who serves whom• Information systems


Shipping Via Cross docksShipping Via Cross docks

• Warehouses: Receiving, Sorting, Storing, Order Picking, Shipping

• Cross Docks = Warehouses without inventory Receiving, Sorting, Shipping

Sorting

Receiving

Inbound shipments

Shipping

Outbound shipments

Requires coordination & IT supportRequires coordination & IT support


Shipping Via Cross docks (Cont’d)Shipping Via Cross docks (Cont’d)

• Goods spend at most 48 hours in the warehouse

• Cross Docking avoids inventory and handling costs,

• Stores trigger orders for products

• Very difficult to manage

• Requires advanced information technology

• All of Wal-Mart’s distribution centers, suppliers and stores are electronically linked to guarantee that any order is processed and executed in a matter of hours



• Goods spend at most 48 hours in the warehouse

• Cross Docking avoids inventory and handling costs,

• Stores trigger orders for products

• Very difficult to manage

• Requires advanced information technology

• All of Wal-Mart’s distribution centers, suppliers and stores are electronically linked to guarantee that any order is processed and executed in a matter of hours



• Wal-Mart operates a private satellite-communications system that sends point-of-sale data to all its vendors allowing them to have a clear vision of sales at the stores

• Needs a fast and responsive transportation system

• Wal-Mart has a dedicated fleet of 2000 truck that serve their 19 warehouses. This allows them to ship goods from warehouses to stores in less than 48 hours replenish stores twice a week on average


Comparing Three StrategiesComparing Three Strategies

W2Factory

W1

W3

Decentralized system: r = 1, L

Safety stock at each warehouse proportional to

Total safety stock

i L 1 3

ii 1

z L 1


Comparing Three Strategies Comparing Three Strategies (Cont’d)(Cont’d)

W2Factory

W1

W3Idealized system: r=1, L

Total safety stock

Safety stock at each warehouse3

2ii3

i 1i

i 1

z L 1

32i

i 1

z L 1


Comparing Three Strategies Comparing Three Strategies (Cont’d)(Cont’d)

W2Factory

W1

W3

Cross-dock system: r = 1, L = L1 + L2

Each period:• System replenishment order = d1 + d2 + d3

• Allocate stock receipts to balance inventories• Transship allocations

L1

L2


Comparing Three StrategiesComparing Three Strategies

Total safety stock

Decentralized

Idealized

Cross-dock

n L 1

n L 1

12

Ln L 1

n

n “identical” warehouses, each with standard deviation of demand =

© Stephen C. Graves 65

L1=0 L1=2 L1=5 L1=8 L1=10 Ideal

n=1 166 166 166 166 166 166

n=2 332 316 292 265 245 235

n=3 497 466 415 357 312 287

n=5 829 766 661 536 433 371

n=10 1658 1517 1275 975 707 524

Safety stock L = 10, = 50, n warehouses with same

Cross-Docking v.s. Idealized SystemCross-Docking v.s. Idealized System

© Stephen C. Graves 66

# of Ws One-echelon Two-echelon

n=1 150 212

n=2 300 342

n=3 450 457

n=4 600 566

n=5 750 671

Decentralized v.s. CDCDecentralized v.s. CDC


Revenue ManagementRevenue Management

• Example:– A cruise ship with C=400 identical cabins– The Price-Quantity relationship



Price

No. seats

2000

1000

P=2000-2Q


RevenueRevenue Management

• Example:– A cruise ship with C=400 identical cabins– The Price-Quantity relationship

• What is the price that the company should charge to maximize revenue?



Price

No. seats

P0=1200

C=500

Revenue=480,000



Price

No. seats

P0=1200

C=400

Money on the Table=160,000



Price

No. seats

P2=1600

Q2=200



Price

No. seats

Q1 =400

P1=1200

Q2=200

P2=1600

Revenue=1600(200) + 1200(400-200)=560,000



• Can we increase revenue more?



Price

No. seats

Q1 =400

P1=1200

Q2=200

P2=1600

P3=1800

Q3=100

Revenue=1800(100) + 1600(200-100) + 1200(400-200)=580,000


How can the firm prevent customers How can the firm prevent customers from moving from one class to from moving from one class to

another?another?

Leisure

Travelers

Business

Travelers

No

Offer

No

Demand

Sensitivity to

Price

Sensitivity to Duration

Sensitivity to Flexibility

High Low

Low

High

Mail-in RebateMail-in Rebate

• To differentiate between customers based on their sensitivity to price

• Adding significant hurdle to the buying process; to receive rebate

• Have to complete and mail the coupon to the manufacturer

• Those customers willing to pay the higher price will not necessarily send the coupon


• A Retailer and a manufacturer.– Retailer faces customer demand.

– Retailer orders from manufacturer.

Selling Price=?

Wholesale Price=$900

Retailer Manufacturer

Variable Production Cost=$200

ExampleExample


Example (Cont’d)Example (Cont’d)

Demand

Price

10000

2000

P=2000-0.2Q


Retailer Expected Profit (No Rebate)

0

200,000

400,000

600,000

800,000

1,000,000

1,200,000

1,400,000

1,600,000

500 1,000 1,500 2,000 2,500 3,000 3,500 3,654 4,110 4,567 4,547

Order

Re

taile

r E

xp

ec

ted

Pro

fit

$1,370,096


Manufacturer Profit (No Rebate)

0

1,000,000

2,000,000

3,000,000

4,000,000

5,000,000

6,000,000

500

1,00

01,

500

2,00

02,

500

3,00

03,

500

3,65

44,

110

4,56

74,

547

4,96

15,

374

5,78

86,

201

6,61

47,

028

7,44

17,

855

Order

Man

ufa

ctu

rer

Pro

fit

$1,750,000


Example (Cont’d)Example (Cont’d)

Demand

Price

10000

2000

P=2100-0.2Q


Retailer Expected Profit ($100 Rebate)

0

200,000

400,000

600,000

800,000

1,000,000

1,200,000

1,400,000

1,600,000

1,800,000

1,000 1,500 2,000 2,500 3,000 3,500 4,000 4,110 4,567 4,547 4,961

Order

Ret

aile

r E

xpec

ted

Pro

fit

$1,644,115


Manufacturer Profit ($100 Rebate)

0

1,000,000

2,000,000

3,000,000

4,000,000

5,000,000

6,000,000

1,00

01,

500

2,00

02,

500

3,00

03,

500

4,00

04,

110

4,56

74,

547

4,96

15,

374

5,78

86,

201

6,61

47,

028

7,44

17,

855

8,26

8

Order

Man

ufa

ctu

rer

Pro

fit

$1,810,392

© Zhi-Long Chen86

Main ReferenceMain Reference

• Presentation slides of Prof. Zhi-Long Chen in the course “Supply Chain Strategies”, MIT, Mass, US.

• Simchi-levi, D., Kaminski, P. and Simchi-levi, E. (2004), Chapter 3, “Inventory Management”.

• Simchi-Levi, D., Kaminsky, P. and Simchi-Levi, E. (2003), “Designing and Managing the Supply Chain”, Irwin McGraw Hill, 2nd edition.

1 logistics & supply chain management strategies peerapol sittivijan...

Documents