Goals - NCAR Research Applications Laboratory | RAL

FDDA Systems Engineering
Training
Introduction (FDDA 101)
February 2008
INTRODUCTION Goals
. Get all FDDA support staff on the same page, e.g., jump onto a
different project and immediately know how to get around and
contribute
. Train for ability to answer user support questions, i.e., troubleshoot
system failures
. Train for ability to setup a new FDDA system
. Train for ability to build and install the FDDA code base on a new
cluster
. Train for ability to customize a model configuration
. Train for ability to extend FDDA capabilities
. Consolidate and extend existing documentation
Ground rules
. DO
< Be curious!
< Attend all sessions
. DON'T
< Bring cell phones
< Bring laptops
< Slam
< Focus on redesign (these are topics for other venues)
Structure
. What will be covered
. Who will be instructing
. Pairing up
. Exercises
. How to get help
< Peers
< Teaching assistants
< System administrators
. Use of the "parking lot"
. FDDA 201: Advanced Topics
DAY 1
Fundamentals
. NWP 101
. Obs nudging
Overview
. FDDA vs. NWP forecasting systems
< Cycling
< Restarts and cold starts
. RT vs. Climo vs. Ensembles
. FDDA inputs and outputs
. Background and history within NSAP
< MM5 twice-daily and FDDA assimilation systems on SGI's
< original "RT-FDDA" started in late 2000 for DPG
< cycle timing defined based on data availability and user output
needs
< original "C-FDDA" was collected in real-time starting in 1997
Resources
. Repositories
< CVS
. Source code
. Scripts
. Templates
. Projects
< Data sources
. Realtime
. Static
. Documentation
. Other useful information
< Mail lists
< Web sites
< MM working group
. Examples of operational systems
Exercise 1
. Verify your login on dev-c1
. Examine web site for an example operational system
. Answer questions about operational cycling
DAY 2
Cluster Basics
. Clusters vs. servers (MAC vs. DAS)
. What is a 'U"?
. Node vs. computer
. Resource allocation (MOAB, Torque, LSF, etc)
. High speed interconnects
. Inter-node parallelization
. Multi-core systems (processors vs. cores)
. Intra-node processing
. Mixed mode: distrib memory / parallel memory
. MPP vs. OpenMP
. Use of individual and project accounts
FDDA Framework (Part I)
. Architectural structure / layers
< csh scripts
< perl scripts
. rtfddaflex systems
. model manager systems
< environment variable settings/definitions
< namelist templates, constants files
Exercise 2
. Setup your individual fdda account on dev-c1
. Answer questions about the cluster hardware
. Examine (do not edit) sample project (ATEC CRTC) on dev-c1
. Answer questions about operational framework Day 3 FDDA Framework (Part II)
. Input data processing
< Initial boundary conditions
< Observations
< Data sources
< Data paths
. Model core
< Choosing a model: MM5 vs. WRF
< Framework differences between MM5 & WRF
. Input required
. Output generated
. Output data processing, aka post-processing
< architectural structure and control
< templates available
< configuration options
< output destinations and configurations
. Known issues?
Building from Source Code
. Cvs hierarchy
. Using the checkout/build script
. Using the new makefiles
. Nightly builds
. Supported platforms
. Testing a new build
. Known issues?
Exercise 3
. Checkout and build in your individual R&D development space on dev-c1
(?)
. Day 4 Project Implementation
. Who: MM5, WRF
. What: Singleton, Ensemble
. When: RT, Climo, Case study
. Where: Domain, resolution, terrain
. Balance: Hardware / cycling / forecast length / application goal
. Establish available input data and path
. Determining output products to generate and where to send them
. Known issues?
Project Validation: Engineering
. Details to be fleshed out
. Known issues?
Project Validation: Scientific
. Subjectively evaluate the model performance by comparing the
model RT output with various data (sat. rad. other op models ...).
Understanding the weather scenarios in the region.
. Monitor the obs data is properly incorporated and look
for special local obs.
. Generate and evaluate model conventional verification statistics
to determine a). How the model generally performs; b). If we should
reduce cold-start cycles from 7 days to 3 days due to too thin data
available in the region.
. Identify model areas/aspects (pro and cons) for improving the model
system in the region. For example, high-res SST of the water bodies in
the region; obvious inconsistent/biased land use specification;
model physics (too much clouds? surface T bias?)...
. Known issues?
Exercise 4 Day 5 Monitoring & Troubleshooting
. Monitoring scripts
. Error logs, status mail, web status pages
. Common failure points
. Routine checklists
. Known issues
< Resource contention on smac-c2
< Others?
CFDDA Details
. Overview
. How it differs from RTFDDA
. Building from source
. Project implementation
. Troubleshooting
. Known issues?
Exercise 5
Day 6 Model Manager
. How to submit jobs
. Queues, quotas, raid partitions
. MOAB portal
. Known issues?
Future Direction
. Regression testing
. E-RTFDDA
. Gridded Bias Correction
. QC Observation data service
. Storage for shared data
. Network topology changes
. Bug tracking
. Suggestions?
Wrap-up
. Questions from Exercises
. Compiling notes and feedback
. What next?
FDDA 201: Advanced topics
. Relocating an FDDA job to a different cluster
. Maintenance & upgrades
< Hardware and OS-level upgrades
< Changes to 3rd party packages and compiler
< New releases from MMM (code, vtables, etc.)
< Changes to input data formats
. Extending FDDA capabilities
< Modifying and installing new code
< Adding new input data streams
< Generating new output products
< NCAR-graphics vs. NCL
. E-RTFDDA
. Purchasing a new project cluster
. Performance issues
. Security concerns -----------------------
10 min
Terrib
2:40-2:50 10 min
Laurie
2:50-3:00 30 min
Laurie
2:10-2:40 30 min
Andrea
1:40-2:10 10 min
Terrib
1:30-1:40 20 min
Exby
1:30-1:50 60 min
Laurie
1:50-2:50 10 min
Exby/Laurie
2:50-3:00 60 min
Laurie
1:30-2:30 20 min
Deirdre
2:30-2:50 10 min
Deirdre
2:50-3:00 50 min
Laurie
1:30-2:20 10 min
Laurie/Andrea
2:50-3:00
10 min
Laurie
2:20-2:30
20 min
Andrea
2:30-2:50
40 min
Andrea
1:30-2:30 40 min
Hank???
1:30-2:30 10 min
Hank/Andrea
2:50-3:00
15 min
Terrib
2:00-2:15 30 min
Elena
1:30-2:00 45 min
Group
2:15-3:00