Experiences running Dynamic Traffic Assignment Simulations at scale using HPC Infrastructure Amit Gupta Joint work with: Weijia Xu Texas Advanced Computing.

Slides:



Advertisements
Similar presentations
Multicast Tree Reconfiguration in Distributed Interactive Applications Pål Halvorsen 1,2, Knut-Helge Vik 1 and Carsten Griwodz 1,2 1 Department of Informatics,
Advertisements

Data and Computer Communications Ninth Edition by William Stallings Chapter 12 – Routing in Switched Data Networks Data and Computer Communications, Ninth.
Advanced Data Structures
Mobile and Wireless Computing Institute for Computer Science, University of Freiburg Western Australian Interactive Virtual Environments Centre (IVEC)
© nCode 2000 Title of Presentation goes here - go to Master Slide to edit - Slide 1 Reliable Communication for Highly Mobile Agents ECE 7995: Term Paper.
1 Advanced Data Structures. 2 Topics Data structures (old) stack, list, array, BST (new) Trees, heaps, union-find, hash tables, spatial, string Algorithm.
Teknik Routing Pertemuan 20 Matakuliah: H0484/Jaringan Komputer Tahun: 2007.
11 Quantifying Benefits of Traffic Information Provision under Stochastic Demand and Capacity Conditions: A Multi-day Traffic Equilibrium Approach Mingxin.
Wireless Sensor Network for Tracking the Traffic in INTERNET Network Routers Part 2 Supervisor:Mark Shifrin Students:Yuri Kipnis Nir Bar-Or Date:30 September.
Spring Routing & Switching Umar Kalim Dept. of Communication Systems Engineering 06/04/2007.
Map Analysis with Networks Francisco Olivera, Ph.D., P.E. Department of Civil Engineering Texas A&M University Some of the figures included in this presentation.
Norman W. Garrick Trip Assignment Trip assignment is the forth step of the FOUR STEP process It is used to determining how much traffic will use each link.
On Scaling Time Dependent Shortest Path Computations for Dynamic Traffic Assignment Amit Gupta, Weijia Xu Texas Advanced Computing Center Kenneth Perrine,
TRIP ASSIGNMENT.
On Self Adaptive Routing in Dynamic Environments -- A probabilistic routing scheme Haiyong Xie, Lili Qiu, Yang Richard Yang and Yin Yale, MR and.
Personalized Ontologies for Web Search and Caching Susan Gauch Information and Telecommunications Technology Center Electrical Engineering and Computer.
Jennifer Rexford Princeton University MW 11:00am-12:20pm Wide-Area Traffic Management COS 597E: Software Defined Networking.
An Empirical Comparison of Microscopic and Mesoscopic Traffic Simulation Paradigms Ramachandran Balakrishna Daniel Morgan Qi Yang Caliper Corporation 14.
Source: NHI course on Travel Demand Forecasting (152054A) Session 10 Traffic (Trip) Assignment Trip Generation Trip Distribution Transit Estimation & Mode.
Distributed Quality-of-Service Routing of Best Constrained Shortest Paths. Abdelhamid MELLOUK, Said HOCEINI, Farid BAGUENINE, Mustapha CHEURFA Computers.
©2009 Proprietary and Confidential DTA in practice: Modeling dynamic networks in the real world Michael Mahut, Ph.D. INRO Montreal, Canada.
© 2014 HDR, Inc., all rights reserved. COUNCIL BLUFFS INTERSTATE SYSTEM MODEL Jon Markt Source: FHWA.
Integration of Transportation System Analyses in Cube Wade L. White, AICP Citilabs Inc.
Regional Traffic Simulation/Assignment Model for Evaluation of Transit Performance and Asset Utilization April 22, 2003 Athanasios Ziliaskopoulos Elaine.
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY San Francisco’s Dynamic Traffic Assignment Model Background SFCTA DTA Model Peer Review Panel Meeting July.
Master’s Thesis Competition Date Monday, 17 th May ‘10.
On Activity-Based Network Design Problems JEE EUN (JAMIE) KANG, JOSEPH Y. J. CHOW, AND WILL W. RECKER 20 TH INTERNATIONAL SYMPOSIUM ON TRANSPORTATION AND.
Comparing Dynamic Traffic Assignment Approaches for Planning
A Survey of Distributed Task Schedulers Kei Takahashi (M1)
David B. Roden, Senior Consulting Manager Analysis of Transportation Projects in Northern Virginia TRB Transportation Planning Applications Conference.
Architecture David Levinson. East Asian Grids Kyoto Nara Chang-an Ideal Chinese Plan.
Dynamic Origin-Destination Trip Table Estimation for Transportation Planning Ramachandran Balakrishna Caliper Corporation 11 th TRB National Transportation.
Network effects from improved traffic signals Kristina Schmidt Transek AB.
Minimax Open Shortest Path First (OSPF) Routing Algorithms in Networks Supporting the SMDS Service Frank Yeong-Sung Lin ( 林永松 ) Information Management.
Mobile Agent Migration Problem Yingyue Xu. Energy efficiency requirement of sensor networks Mobile agent computing paradigm Data fusion, distributed processing.
A Dynamic Traffic Simulation Model on Planning Networks Qi Yang Caliper Corporation TRB Planning Application Conference Houston, May 20, 2009.
DVRPC TMIP Peer Review Dynamic Traffic Assignment (DTA) Oct. 29 th, 2014.
TRB Planning Applications May 2009, Houston,TX Changing assignment algorithms: the price of better convergence Michael Florian and Shuguang He INRO.
Methodological Considerations for Integrating Dynamic Traffic Assignment with Activity-Based Models Ramachandran Balakrishna Daniel Morgan Srinivasan Sundaram.
Analysis of the IH 35 Corridor Through the Austin Metropolitan Area TRB Planning Applications Conference Jeff Shelton Karen Lorenzini Alex Valdez Tom Williams.
Teknik Routing Pertemuan 10 Matakuliah: H0524/Jaringan Komputer Tahun: 2009.
COS 420 Day 15. Agenda Finish Individualized Project Presentations on Thrusday Have Grading sheets to me by Friday Group Project Discussion Goals & Timelines.
Scalable and Topology-Aware Load Balancers in Charm++ Amit Sharma Parallel Programming Lab, UIUC.
Ephemeral Network Broker to Facilitate Future Mobility Business Models/Transactions A collaboration between Ford University Research Program and University.
Data Consolidation: A Task Scheduling and Data Migration Technique for Grid Networks Author: P. Kokkinos, K. Christodoulopoulos, A. Kretsis, and E. Varvarigos.
Urban Traffic Simulated From A Dual Perspective Hu Mao-Bin University of Science and Technology of China Hefei, P.R. China
1 Traffic Engineering By Kavitha Ganapa. 2 Introduction Traffic engineering is concerned with the issue of performance evaluation and optimization of.
September 2008What’s coming in Aimsun: New features and model developments 1 Hybrid Mesoscopic-Microscopic Traffic Simulation Framework Alex Torday, Jordi.
William Stallings Data and Computer Communications
Network Layer COMPUTER NETWORKS Networking Standards (Network LAYER)
Route Choice Lecture 11 Norman W. Garrick
Data Center Network Architectures
Prepared for 16th TRB National Transportation Planning Applications Conference Outline Gap Value in Simulation-Based Dynamic Traffic Assignment (DTA) Models:
Calculating time-dependent travel times for VRPs
The Underlying Technologies
1st November, 2016 Transport Modelling – Developing a better understanding of Short Lived Events Marcel Pooke – Operational Modelling & Visualisation Manager.
Parallel Programming By J. H. Wang May 2, 2017.
Network Assignment and Equilibrium for Disaggregate Models
Jim Henricksen, MnDOT Steve Ruegg, WSP
Frank Yeong-Sung Lin (林永松) Information Management Department
Routing and Logistics with TransCAD
Degree-aware Hybrid Graph Traversal on FPGA-HMC Platform
COS 561: Advanced Computer Networks
Data and Computer Communications
Route Choice Lecture 11 Norman W. Garrick
Backbone Traffic Engineering
Frank Yeong-Sung Lin (林永松) Information Management Department
2019/5/13 A Weighted ECMP Load Balancing Scheme for Data Centers Using P4 Switches Presenter:Hung-Yen Wang Authors:Peng Wang, George Trimponias, Hong Xu,
2019/9/14 The Deep Learning Vision for Heterogeneous Network Traffic Control Proposal, Challenges, and Future Perspective Author: Nei Kato, Zubair Md.
CSCI 465 Data Communications and Networks Lecture 16
Presentation transcript:

Experiences running Dynamic Traffic Assignment Simulations at scale using HPC Infrastructure Amit Gupta Joint work with: Weijia Xu Texas Advanced Computing Center Kenneth Perrine, Dennis Bell, Natalia Ruiz-Juri Network Modeling Center Center for Transportation Research

Dynamic Traffic Assignment Used to Model complex interactions between –Traveller   Traveller –Traveller   Road Infrastructure Usually Simulation Based Shortest Path calculations –Used to model traveller behavior to congestion –Intensive part of the process –Time dependent Traffic conditions change 1

Road Networks & Travel Time Turn Movement Delays Signal Delays 2 Prohibitions –(One Ways, Closed Roads) Tolls, Vehicle Types * Time Dependent *

3 Transportation: Road Networks

Dynamic Traffic Assignment Static Traffic Assignment –Inflow = Outflow Dynamic Traffic Assignment –More realistic modeling of congestion Lambda –Fraction of vehicles that change route Equilibrium –For each O/D pair, all routes experience the same travel time. Different route choice cant lower cost. Gap - Measure of proximity to equilibrium 4

DTA Workflow 5

Our Approach Example Profile VISTA –DTA and Simulation Framework –Popular in Transportation Research “strace –c” Cross verified with other profiling (iostat) 6

Example Runtime Profile I/O Bottleneck –Initial profiling Upto 77% time spent here –Primarily Route File activity Moved target IO to local Ramdisk –Most I/O ops in-memory 7

Our Approach Computational Bottleneck –Profiling indicated TDSP( Update Labels ) is the main computational Load 8

TDSP Is a “Label Correcting” algorithm Has 2 main parts –Update Labels Incorporates the activity of most recent simulation Changes the cost (label) values at nodes Causes upstream costs to change Maintains pointers to downstream nodes –Search for Routes Back traces the downstream pointers Constructs the shortest path topology 9

TDSP 10 Origin Destination

Update Labels 11 Origin Destination

TDSP Challenges For precision –Larger number of labels maintained –Implies Computational overhead Simulation activity changes travel time –Label values keep changing –Shortest paths also change –Stored on file and looked up –Small, Random reads. I/O overhead 12

13

Update Labels: Queuing Scheme Scan Eligible List –All Label Correcting algorithms use it –Intermediate structure to hold nodes for later VISTA used Dequeue scheme –“Visited nodes” inserted to the front –No prioritization otherwise –Order of Nodes depends on order in which Links are visited –“Feedback effect” during label updates 14

15

Priority Queue Order of Node evaluation matters Prioritize Nodes during evaluation –Use the first time label –Nodes label values converge faster –Reduces overall time for Update Labels => TDSP finishes faster –Average case is improved Already congested link stays that way –Analysis periods like “Rush Hour” time windows –Common for many research investigations 16

17

18

Datasets NetworkNodesLinksO/D PairsDemand (Vehicles) Downtown Austin South Austin Central Austin Austin Region AM Austin Region PM

20

21

Summary Large performance gains (6x-12x) –Relatively minor code modifications –Changes to application architecture Other things attempted –Using Accelerators (Xeon Phi) Offload calculations at every hop Overhead is too high –Different Label for priority Results similar Depends on demand 22

Other code in workflow Update cost parallelization –Runs a different test for different lambda values –Used in analysis –Test data structure Deep Tree like hierarchy –Simulation Results »Graph topology »… –File Store –… Not trivial to serialize Simple solution using MPI_Scatter/Gather not possible 23

Future Work Web portal to kick off analysis jobs on TACC infrastructure Examine possibility of serialization for lambda test related data structures Architectures moving away from offload model –Determine tradeoffs in Knights Landing Investigate using Hadoop/Spark based graph engines for more scalable solutions 24

Acknowledgement Current –XSEDE ECSS –NSF Stampede Supercomputer Previous Funding –Texas Department of Transportation –CAMPO Capital Area Metropolitan Planning Organization Questions ?... 25