Interstate Data Moving and the Last Block Problem: Lessons Learned in the CAPS Spring Experiment 2014 Keith A. Brewster, Ph.D. Center for Analysis and.

Slides:



Advertisements
Similar presentations
Oklahoma Center for High Energy Physics A DOE EPSCoR Proposal Submitted by a consortium of University of Oklahoma, Oklahoma State University, Langston.
Advertisements

Kathy Benninger, Pittsburgh Supercomputing Center Workshop on the Development of a Next-Generation Cyberinfrastructure 1-Oct-2014 NSF Collaborative Research:
User Perspective: Using OSCER for a Real-Time Ensemble Forecasting Experiment…and other projects Currently: Jason J. Levit Research Meteorologist, Cooperative.
1 PSC update July 3, 2008 Ralph Roskies, Scientific Director Pittsburgh Supercomputing Center
Interstate Data Moving and the Last Block Problem: Lessons Learned in the CAPS Spring Experiment 2014 Keith A. Brewster, Ph.D. Center for Analysis and.
NETWORKING HARDWARE.
James Deaton Chief Technology Officer OneNet Glass, Graphs and Gear.
Jack Kain Hazardous Weather Forecasts & Warnings Jack Kain Hazardous Weather Forecasts & Warnings Science Infusion 1.
Atmospheric Data Analysis on the Grid Kevin Hodges ESSC Co-workers: Brian Hoskins, Lennart Bengtsson Lizzie Froude.
Integration of Storm Scale Ensembles, Hail Observations, and Machine Learning for Severe Hail Prediction David John Gagne II Center for Analysis and Prediction.
GOES-R Proving Ground Activities at the NWS Storm Prediction Center Dr. Russell S. Schneider Chief, Science Support Branch NWS Storm Prediction Center.
L inked E nvironments for A tmospheric D iscovery leadproject.org Using the LEAD Portal for Customized Weather Forecasts on the TeraGrid Keith Brewster.
GOES-R Proving Ground Storm Prediction Center and Hazardous Weather Testbed Chris Siewert 1,2, Kristin Calhoun 1,3 and Geoffrey Stano OU-CIMMS, 2.
HWT Spring Forecasting Experiment: History and Success Dr. Adam Clark February 25, 2015 National Weather Center Norman, Oklahoma.
THORPEX Interactive Grand Global Ensemble (TIGGE) China Meteorological Administration TIGGE-WG meeting, Boulder, June Progress on TIGGE Archive Center.
The II SAS Testbed Site Jan Astalos - Institute of Informatics Slovak Academy of Sciences.
NETWORK HARDWARE CABLES NETWORK INTERFACE CARD (NIC)
The Internet The History and Future of the Internet.
Sponsored by the National Science Foundation A New Approach for Using Web Services, Grids and Virtual Organizations in Mesoscale Meteorology.
NOAA Hazardous Weather Test Bed (SPC, OUN, NSSL) Objectives – Advance the science of weather forecasting and prediction of severe convective weather –
INFORMATION EXTRACTION AND VERIFICATION OF NUMERICAL WEATHER PREDICTION FOR SEVERE WEATHER FORECASTING Israel Jirak, NOAA/Storm Prediction Center Chris.
Term 2, 2011 Week 2. CONTENTS Communications devices – Modems – Network interface cards (NIC) – Wireless access point – Switches and routers Communications.
R2O and O2R between NSSL and SPC: The benefits of Collocation Jack Kain, Steve Weiss, Mike Coniglio, Harold Brooks, Israel Jirak, and Adam Clark.
 A hub is a central connecting device in a network.  Each node is connected directly to the hub.  They receive a data packet from one node and send.
CAPS Realtime 4-km Multi-Model Convection-Allowing Ensemble and 1-km Convection-Resolving Forecasts for the NOAA Hazardous Weather Testbed (HWT) 2009 Spring.
Activity 1 5 minutes to discuss and feedback on the following:
OKLAHOMA Supercomputing Symposium 2011 University of Oklahoma October 11, 2011 James Wicksted, RII Project Director Associate Director, Oklahoma EPSCoR.
Slide 1 E-Science: The Impact of Science DMZs on Research Presenter: Alex Berryman Performance Engineer, OARnet Paul Schopis, Marcio Faerman.
High Resolution Assimilation of Radar Data for Thunderstorm Forecasting on OSCER Keith A. Brewster, Kevin W. Thomas, Jerry Brotzge, Yunheng Wang, Dan Weber.
CAPS Realtime 4-km Multi-Model Convection-Allowing Ensemble and 1-km Convection-Resolving Forecasts for the NOAA Hazardous Weather Testbed (HWT) 2009 Spring.
Assignment 3 Jacob Seiz. Hub A hub provides a central access point for a network. Through multiple I/O ports a hub can connect multiple Ethernet devices.
Network and Server Basics. Learning Objectives After viewing this presentation, you will be able to: Understand the benefits of a client/server network.
Research Focused Networking ● Science DMZ ● Network segment topologically close to campus edge, available as a co-location space for research data ● Optimized.
UNM SCIENCE DMZ Sean Taylor Senior Network Engineer.
1.4 wired and wireless networks lesson 2
Bryan Learn Pittsburgh Supercomputing Center
Transfer of data in ICT systems
Local Area Networks Honolulu Community College
Tuning an Analysis and Incremental Analysis Updating (IAU) Assimilation for an Efficient High Resolution Forecast System Keith A. Brewster1
A few examples of heavy precipitation forecast Ming Xue Director
Multi-Scale Spatiotemporal Mining of Atmospheric Data
Chapter 3 Computer Networking Hardware
Tuning an Analysis and Incremental Analysis Updating (IAU) Assimilation for an Efficient High Resolution Forecast System Keith A. Brewster1
TEIN update – APAN44 Douglas Harré Chief Engagement Officer
Service Challenge 3 CERN
Hydrometeorological Predication Center
CAPS is one of the first 11 NSF Science and Technology (S&T) Centers
Southwest Tier 2 Center Status Report
Center for Analysis and Prediction of Storms (CAPS) Briefing by Ming Xue, Director CAPS is one of the 1st NSF Science and Technology Centers established.
High-resolution air quality forecasting for Hong Kong
3.2 Virtualisation.
An Overview of the ITTC Networking & Distributed Systems Laboratory
Welcome To : Group 1 VC Presentation
High resolution radar data and products over the Continental United States National Severe Storms Laboratory Norman OK, USA.
Keith A. Brewster1 Jerry Brotzge1, Kevin W
Grid Canada Testbed using HEP applications
CAPS Mission Statement
An overview of real-time radar data handling and compression
Computer networking In this section of notes you will learn the rudiments of networking, the components of a network and how to secure a network.
Introduction to Research Facilitation
Firewalls Routers, Switches, Hubs VPNs
Smart Water Fund Products
CAPS Real-time Storm-Scale EnKF Data Assimilation and Forecasts for the NOAA Hazardous Weather Testbed Spring Forecasting Experiments: Towards the Goal.
Henry Neeman, University of Oklahoma
Cloud computing mechanisms
IP Control Gateway (IPCG)
MOGREPS developments and TIGGE
- Measurement of traffic volume -
Computer Networks Lesson 2.
Securing HTCondor Flocking
Presentation transcript:

Interstate Data Moving and the Last Block Problem: Lessons Learned in the CAPS Spring Experiment 2014 Keith A. Brewster, Ph.D. Center for Analysis and Prediction of Storms University of Oklahoma

SPC/NSSL Spring Program in the Hazardous Weather Testbed Testing and calibration of new forecasting methods in a simulated operational setting 6 weeks in spring season Collaboration among NOAA research units NOAA operational units Universities Private sector Testbed located between the NOAA Storm Prediction Center and Norman National Weather Service Forecast Office

CAPS Spring Experiment Part of NOAA/SPC Spring Experiment at the Hazardous Weather Testbed Run Large Ensemble of Convection-Allowing NWP Forecasts for 6-weeks in Spring New methods for severe weather prediction in 1-2 days time frame 25 NWP models run at XSEDE Centers 2013-2014 Darter at NICS (UTenn @ Oak Ridge)

Goal: “Real-time” 4D Data Visualization Procedure Since 2007 Run models at PSC or NICS Bring 2D files and images back 2D fields and levels pre-selected Issues Does not allow full 3D visualization May want to examine other fields and levels Need to move 3D Data from NICS at UTenn CAPS at OU

Scoping the Task CONUS Domain at 4-km Resolution 1163 x 723 x 53 Output for one time: 4.2 GB Domain decomposition onto 384 (6x64) processors results in 384 split files per output time, each file = 11 MB For smooth animations, 10-minute output is generated for 5 members covering the afternoon and evening, forecast hours 18-30. Complete Forecast: 60-h forecast, hourly output + 10 minute output 18h-30h: 121 output times, 508 GB Day-1 Afternoon and Evening for Animation Forecast 18h-30h with 10-minute output: 73 output times, 307 GB

Plan “A” Workflow WRF Model wrfout split files NICS Darter GridFTP or bbcp Daily Region of Interest joinWRF Join and subset wrfout joined files CAPS Server WDSS-II Converter vdfcreate WDSS-II files VAPoR vdf files Each full split file: 112MB sftp or scp sftp or scp VAPoR NSSL or SPC Machine in HWT WDSS-II

Data Route NICS/U Tenn in Oak Ridge, Tennessee to CAPS/OU in Norman, Oklahoma 845 miles 13 hours via MichelinNet Map from mapquest.com

Internet Route

Internet2

Internet Route “Last Mile” OneNet Tulsa to OU – Norman (4PP) 4PP to National Weather Center Across the parking lot National Weather Center to CAPS Switch to File Server System to CAPS Office Workstation via Firewall

Recent Networking Initiatives University of Tennessee BLAST OneOklahoma Friction Free Network (OFFN) National Weather Center Upgrade University of Tennessee BLAST 100 Gps Upgrade of Research Network Includes connections to HPC at NICS

Recent Networking Initiatives OneOklahoma Friction Free Network (OFFN) NSF Campus Cyberinfrastructure-Network Infrastructure and Engineering Program (CC-NIE) Establish 10 Gbps Network Ring OU-OSU-Langston-Tandy Supercomputing Ctr

Traffic & Last Block Problem Testing revealed a “last block” problem, actually within the building itself, mostly due to a slow firewall.

Packing and Compression? Try creating compressed tar file before sending? Sending large files is faster ~100 MB/s vs ~10 MB/s BUT! Creating a compressed tar file takes time Test for 1-hour of Full Domain Split Files Operation Time bbcp individual split files 50 min With Compression Without Compression Operation Time $ tar –zcvf 22 min bbcp 2.5 min Total 24.5 min Operation Time $ tar -cvf 15 min bbcp 5 min Total 20 min

Plan “D” Workflow Daily Region of Interest NICS Darter joinWRF Join and subset WRF Model wrfout split files wrfout joined files bbcp SPC Workstation CAPS Stratus WDSS-II Converter sftp wrfout joined files WDSS-II files vdfcreate VAPoR vdf files New Merged file size: 112 MB Transfer speeds: NICS to Boomer scp ~50 Mbytes/s Large file: NICS to data1.oscer.ou.edu bbcp ~70-100 Mbytes/s Boomer to CAPS: 5 Mbytes/s sftp CAPS Laptop WDSS-II VAPoR

Re-Scoping the Task Selected Subdomain of the Day 203 x 203 x 53 Output for one time: 111.5 MB Single file for each time = 111.5 MB vs 4.2 GB Day-1 Afternoon and Evening for Animation Forecast 18h-30h with 10-minute output: 73 output times, 8.2 GB vs 307 GB Processing and Transfer: ~20 min

Fruits of Labor WDSS-II

Fruits of Labor VAPoR

Lessons Learned Involve networking pros early Your mileage (throughput) may vary Be flexible with workflows Evaluate overhead of all steps Find ways to fund equipment upgrades where needed – slowest link sets your rate Software Programmable Networks/Science DMZ’s may be needed for the largest jobs

Future Plans National Weather Center (September, 2014) Upgrading Network to 10 Gps Switches Software Programmable Networking Enabled Replacing Slow Firewall CAPS (September-October, 2014) Upgrading to Two 10 Gps Gateway Servers with Virtual Router Redundancy Protocol (VRRP) Upgrading servers to 10 Gps network interfaces For 2015 Set Jumbo Size Maximum Transmission Unit (MTU) for route Explore use of Science DMZ within OneNet

Questions? Thanks to: Chris Cook, CAPS Kevin W. Thomas, CAPS Victor Hazlewood and his networking team, UTenn Matt Runion & James Deaton, OneNet Henry Neeman & OSCER Team Mike Coniglio, NOAA/NSSL Contact Info: Keith Brewster CAPS/University of Oklahoma kbrewster@ou.edu