Optimization with Grid Computing

1
Optimization with Grid Computing

• Heinrich Braun

2
Agenda
Introduction Supply Chain Management
Mastering the Algorithmic Planning Complexity
by Grid Computing
Beyond Enterprise wide Optimization Optimizing
the Collaboration
3
Introduction Supply Chain Management
4
Challenge for Standard Software Provider
SAPGeneric Optimizer

• Generic and Best of Breed
• planning level
• vertical industries
• run time requirement
• model complexity (size, constraints, objectives)
• Generic Model (-gt planning level)
• aggregated planning (LP / MILP)
• detailed planning (scheduling)
• Customization (-gt vertical industries)
• specialization the generic model to customer
  problem
• scripting the strategies (decomposition, goal
  programming)
• Scalability (-gt run time)
• greedy versus complex optimizations strategies
• Parallelization by GRID Computing

5
Agenda
Introduction Supply Chain Management
Mastering the Algorithmic Planning Complexity
by Grid Computing
Beyond Enterprise Wide Optimization Optimizing
the Collaboration
6
How to deal with planning complexity? I

• Basic idea Hierarchy of relaxations
• Relaxations are derived by Aggregation
• Time ? Periods
• Product ? Product groups
• (e.g. ignore country specific documentation
  in packaging a product)
• Resource ? Resource Families
• (e.g. summarize similar resources into one
  resource with cumulative capacity)
• Locations ? Regions
• (e.g. aggregate different locations into a
  transportation zone (postal code areas)

7
How to deal with planning complexity? II

• Basic idea Local Search Decomposition
• Global versus local optimality
• Local optimality depends on neighborhood
• High solution quality by local optimization
• Decomposition strategies
• SNP time, resource, product, procurement
• DS time, resource

8
Time Decomposition
9
Time Decomposition
10
Time Decomposition
11
Time Decomposition
12
Time Decomposition
13
Time Decomposition
14
Time Decomposition
15
Leading Edge of Customer Problems
16
Challenge Customer of Food Industry
Demand
22 Distribution Centers
Large model size high complexity 4 million
decision variables
19 Warehouses
150 000 Stock Keeping Units
110 Plants
17
How to deal with planning complexity? III

• Basic idea Local Search
• Global versus local optimality
• Local optimality depends on neighborhood
• High solution quality by local optimization
• Local Optimization Decomposition
• Decomposition strategies
• SNP time, resource, product, procurement
• DS time, resource
• GRID Computing
• Parallelization by local search agents

18
Grid Architecture
Grid-Enabled SAP Applications
Grid Services
Application Deployment
Grid Services Registry
Resource Management
Grid Management
Blade
PC-Cluster
19
Grid Computing Next Generation of Scalability
20
Summary - Mastering the Algorithmic Complexity

• Grid Services Reduce TCO (total cost of
  ownership)
• Adminstration (standard based WebServices /
  NetWeaver Platform)
• Hardware (scalable PC Clusters / Blades)
• SCM Optimizer - Leverage the power of Grid and
  Decomposition
• Faster Response Time (Distribution of Processing)
• Larger Problem Size (Distribution of Main Memory)
• Higher Modelling Complexity
• ? Ready for Next Generation of Optimization
  Problems
• Leading edge of performance for Big Business
• Low Cost for Midsize Business

21
Agenda
Introduction Supply Chain Management
Mastering the Algorithmic Planning Complexity
by Grid Computing
Beyond Enterprise Wide Optimization Optimizing
the Collaboration
22
InCoCo Project (consortium supported by EU
funding)

• InCoCo
• Innovation, Coordination and Collaboration
• in a Service Driven Supply Chain
• Goal Innovative planning methods and
  collaboration policies
• Decomposition of supply chains by autonomously
  planning partners
• Handling heterogenous supply chains (no central
  planning instance)

23
As-Is Situation and Vision
Supplier
Buyer
Win-Win
APS
APS
Purchase order quantities
schedule
Purchase order quantities
schedule

• Current collaborative planning solutions aim
  primarily at the effective communication of
  demand along the inter-organizational supply
  chain.

• Vision Balancing plans from an
  inter-organizational perspective to avoid
  redundant costs increases competitiveness of the
  whole SC

24
Preconditions for Collaboration

• Political dimension
• No disclosure of sensitive data
• Technical dimension
• Negotiation protocol and mapping of products
  needed
• Fast computation of mutual benificial proposals
  by Advanced Planning Systems

25
History

• Concept Car
• excellent research work of the Team Prof.
  Stadtler (Uni Hamburg)
• 2003 Management Strategic Innovation Prize der
  GOR (G. Dudek)
• 2004 Dissertation prize of Gesellschaft
  Operations Research (G. Dudek)
• Model and Validation by G. Dudek
• Two partner relation
• Small model with few products
• Mid term planning level (Linear Program)
• Research Focus of SAP for InCoCo
• Extend the model and negotiation schemas
• Realistic Supply Chain Models
• Based on our benchmark suite of real customer

26
Test Results (Dudek, Stadtler 2005)
27
Research Focus Collaborative Planning

• Negotiaton schema for aligning decentrally
  generated plans
• using only uncritical information (demand and
  supply quantities)
• with a solution quality comparable to globally
  optimized plans
• with only few effort (e.g. 5 iterations)
• based on todays Advanced Planning Systems
• Service Oriented Architecture
• Information Hiding (!!)
• Master Slave Concept
• Master only Mediator for Collaboration
• Indepedent slaves APS-Systems of Supplier and
  Producer
• Proposing improvement potentials of their plans
  with benefit value
• Acceptance of plan changes depends on benefit
  extra costs

28
Summary

• Key Features of SAP software architecture
• Platform Concept (SAP Netweaver)
• Service Oriented Architecture (SOA)
• Grid Computing for Optimization
• Current Enterprise Wide Optimization
• Enabling all Optimizer for the power of
  parallelization
• Faster (solution time), larger (planning models)
• Easier administration of the optimizer server
• Future Optimization of the collaboration of
  supply chain partners
• improving their overal supply chain planning
  solution
• By decentralization for these collaboration
  processes
• less integration and implementation costs
• easier adoption of new partners

29

• Thank you for your attention