Reducing Model Run Times Using Distributed Computing

1

• Reducing Model Run Times Using Distributed
  Computing
• An East-West Gateway COG Case Study

Matthew Martimo, Lubna Shoaib, Victor Siu
2
Model

• Singly Constrained Four Step Model
• Joint Mode Choice-Destination Choice Step
• 2527 TAZ, 68 External Stations
• 8,144 CLM 18,509 Lane Miles
• 2.4 Million Residents
• 31 Transit Modes
• Entire Model in Cube

3
Run Time Issues

• Long Model Run Times 69 hours
• Solution CUBE Cluster
• Multi-Step Distributive Processing
• Independent processing sent to independent
  computers
• Up to 50 reduction in time by adding another
  machine
• Intra-Step Distributive Processing
• Run Times when processing complex calculations
  may be reduced to as little as 100/Number of
  Computers
• Negative savings may be observed with very simple
  calculations
• Estimated Run Time RN (R1-C)/N C
• For St. Louis AM Assignment C 14 min (12)

4
Setup Commander

• Dynamic Number of Nodes Controlled in Application
  Manager
• User selects
• Whether or Not to Distribute
• How Many Nodes will be Used

5
Cluster Run Times
Time (hrmin)
6
Assignment Application
Time (hrmin)
MDP Multi Distt Processing IDP Intra Distt
Processing
7
Matrix Intra-Step Processing
8
Highway Intra-Step Processing
9
Run Time -Assignment
10
St. Louis Model Results