Transcript
Page 1: Region-Scale Evacuation Modeling using GPUs

Region-Scale Evacuation Modeling using GPUs

Towards Highly Interactive, GPU-based Evaluation of Evacuation Transport Scenarios at State-Scale

Kalyan S. Perumalla, PhD Brandon G. Aaby Srikanth B. Yoginath Sudip K. Seal, PhD

National Evacuation ConferenceNew Orleans, LA

Feb 5, 2010

Page 2: Region-Scale Evacuation Modeling using GPUs

2 Managed by UT-Battellefor the U.S. Department of Energy National Evacuation Conference 2010

RealSim Program at ORNL

Real-Time Data Driven Simulation System• Kalyan Perumalla, Ph.D. (PI)

• Brian Worley, Ph.D. (Director)

• Vladimir Protopopescu, Ph.D.

• Srikanth Yoginath

• Sudip Seal, Ph.D.

• Brandon Aaby

• Clayton Thurmer

• Other collaborators

Page 3: Region-Scale Evacuation Modeling using GPUs

3 Managed by UT-Battellefor the U.S. Department of Energy National Evacuation Conference 2010

Evacuation Analysis: New

Capabilities

Real-time or Faster Speed

Accuracy and

ComplexityState-or Regional

Spatial Scale

Our FocusNext Generation Computational Capability

Millions of Grid Cells

GPUs

Field-based Mobility Modeling

Page 4: Region-Scale Evacuation Modeling using GPUs

4 Managed by UT-Battellefor the U.S. Department of Energy National Evacuation Conference 2010

Affordable, off-the-shelf

Desktop-based

GPU: Graphical Processing Units

Example: NVIDIA

• Original GeForce 6000, 7000, 8000, and 9000 series

• Latest GTX 295

• Upcoming GTX 300

CPUCPU

GPUGPU DisplayDisplayMemoryMemory

Page 5: Region-Scale Evacuation Modeling using GPUs

5 Managed by UT-Battellefor the U.S. Department of Energy National Evacuation Conference 2010

Graphics as Computation

Programmable graphics primitives

E.g., pixel shading such as bump mapping and patterned texture mapping

Page 6: Region-Scale Evacuation Modeling using GPUs

6 Managed by UT-Battellefor the U.S. Department of Energy National Evacuation Conference 2010

A Few other Popular Instantiations

Commercial Offerings

• IBM Cell Processor

• NVIDIA GeForce, GTX

• NVIDIA Tesla

Supercomputing Scale

• LANL RoadRunner

Page 7: Region-Scale Evacuation Modeling using GPUs

7 Managed by UT-Battellefor the U.S. Department of Energy National Evacuation Conference 2010

Vertical Grid ElementsIncoming Outgoing

Page 8: Region-Scale Evacuation Modeling using GPUs

8 Managed by UT-Battellefor the U.S. Department of Energy National Evacuation Conference 2010

Horizontal Grid Elements

Incoming Outgoing

Page 9: Region-Scale Evacuation Modeling using GPUs

9 Managed by UT-Battellefor the U.S. Department of Energy National Evacuation Conference 2010

FP

FP

FPFP

Texture Memory

FP=Fragment Processor

Typical Single PhaseExecution on GPU

FP

FP

FP

FP

Texture Memory

FP=Fragment Processor

FP

FP

FPFP

Texture Memory

FP=Fragment Processor

FP

FP

FPFP

Texture Memory

FP=Fragment Processor

FPFPFPFP

Texture Memory

FP=Fragment Processor

FP

FP

FP

FP

Texture Memory

FP=Fragment Processor

v vv

GARFIELD Configurable Display with

User Interaction

Our Multi-phase Execution Framework for Model Flexibility

FP

FP

FPFP

Texture Memory

FP=Fragment Processor

v

Field Model Execution Framework on GPU

Page 10: Region-Scale Evacuation Modeling using GPUs

10 Managed by UT-Battellefor the U.S. Department of Energy National Evacuation Conference 2010

GPU Data Structures and Execution

Page 11: Region-Scale Evacuation Modeling using GPUs

11 Managed by UT-Battellefor the U.S. Department of Energy National Evacuation Conference 2010

System Functional Architecture

Induced Graph Field Generator

Road NetworkDatabase (E.g.,

Tiger)

Field GridResolution (GPUconfiguration)

Texel GridGenerator

DestinationLocations

Sources andInitial LoadingDistributions

InteractiveSpecifications

GPU-basedExecution Engine

QueueingModel(Code)

RuntimeDisplay andVisualizatio

n

CongestionModel(Code)

CustomizedEncoding,

Coloring andAnimation (Code)

GeographicalRegions of

Interest (Latitude,Longitude)

Map ImagesServer orDatabase

Statistics and Reports

1 1

Page 12: Region-Scale Evacuation Modeling using GPUs

12 Managed by UT-Battellefor the U.S. Department of Energy National Evacuation Conference 2010

Louisiana

Page 13: Region-Scale Evacuation Modeling using GPUs

13 Managed by UT-Battellefor the U.S. Department of Energy National Evacuation Conference 2010

Lousiana (continued)

Page 14: Region-Scale Evacuation Modeling using GPUs

14 Managed by UT-Battellefor the U.S. Department of Energy National Evacuation Conference 2010

Texas

Page 15: Region-Scale Evacuation Modeling using GPUs

15 Managed by UT-Battellefor the U.S. Department of Energy National Evacuation Conference 2010

Texas (continued)

Page 16: Region-Scale Evacuation Modeling using GPUs

16 Managed by UT-Battellefor the U.S. Department of Energy National Evacuation Conference 2010

Example Timing Results

Texture Evac Time

Run Time

State Nodes Links X×X Hours Sec

DC 9,559 14,884 1048576 35.20 54.90

LA 413,574 988,458 4194304 65.07 409.59

TN 583,484 1,335,586 3211264 157.91 353.89

FL 1,048,506 2,629,268 4194304 179.20 611.83

TX 2,073,870 5,116,492 3211264 217.60 777.65

Page 17: Region-Scale Evacuation Modeling using GPUs

17 Managed by UT-Battellefor the U.S. Department of Energy National Evacuation Conference 2010

New Capabilities

First multi-million node/link road network simulation

First GPU-based execution of road network simulation

First field-based mobility formulation

Page 18: Region-Scale Evacuation Modeling using GPUs

For additional information

and animations,

visit our

RealSim webpage at

www.ornl.gov/~2ip


Top Related