CCSM Software Engineering

Community Climate System Model www.ccsm.ucar.edu 1CCSM

CCSM Software Engineering

CCSM Software Engineering GroupModel Porting and Performance.Documentation and Training.Software Engineering Practices.

Lawrence BujaCCSM Software Engineering Group


CSEG

Purpose

Core group responsible for CCSM software engineering development

PositionsCSEG ManagerCSEG Quality Assurance LeadCCSM Component Liaisons

CCSM Software Engineering Group


CSEG Member PositionTony Craig CSEG ManagerLawrence Buja QA LeadErik Kluzek ATM LiaisonNancy Norton OCN LiaisonMariana Vertenstein LND LiaisonJulie Schramm ICE LiaisonBrian Kauffman CPL Liaison


Role of CCSM SE ManagerSoftware decision making within CCSM and single point of contact for all CCSM SE Issues

"allow the scientists to do science"

• Prioritize and assign SE tasks• Track activities across CCSM Groups • Coordinate internal and external developers• Model release coordination and closure• Represent SE interests/issues on SSC• Create and Maintain SE plans• Schedule SE training


CCSM PerformanceOCN

ICE

LND

ATM

CPL

8 3 7

8 37

7 4

3 7 2 14

41secs

53 secs/day total (~130 years/month)

NERSC T42/gx1v3 (128Pes)

32

8

8

32

48

Critical path


CCSM Ports

CCSM currently being tested/run on:• NCAR IBM SP• NCAR SGI Origin 2000• NCAR Compaq ES-40• LANL SGI Origin 2000• NERSC IBM SP


CCSM DocumentationCCSM Documentation Project:

- User and Developers Guide for each component

- Quick Start and User Guide for CCSM

- Internal Interface documentation

- Growing online library of CCSM SE documents


CCSM TrainingQuick-start tutorial June 2001

- Half-day tutorial on building and running the CCSM

Week long workshop Fall 2002

- Theory and practice of the CCSM


XP CX1 SEI/CMMExtreme Construx CapabilityProgramming CXone Maturity Model

Modern SE Practices

• CCSM becoming increasing complex• Many new players in the development• Target architectures continue to evolve

Software Engineering Models

Why?

Too Light Just Right Too Heavy


SE Coordination Plan

• Change Review Board (CRB) • Repository • Testing • Status Accounting • Documents • Planning and Prioritization • Management • Other Issues


Roadmap to JuneDec Jan Feb Mar Apr May Jun

CCSM2 Release

Work Shop

CCSMFreeze

CCSM2 Documentation

CPL6 beta

PaleoCCSM

CCSM2 Experiments

CCSM2 Control Runs


Fini


Roadmap to June

CCSM2 beta

CCSM2 alpha tests

CCSM2

Freeze

Oct Nov Dec Jan

CCM

CSIM

POPLSM

Components, interfaces, gridsshould be close to final state

SSCOct 17

NASA CAN due Nov 20

NASA CAN Selections Feb 5


Freezing CSM-2

• Finalizing Component Configurations• Component Acceptance• Validation Procedure


CCSM Issues

•Code Development Process (this meeting)• Who defines process?• Who coordinates the efforts?

• Integration with ACPI and NASA CAN efforts• Insufficient Software Education• Implementation of Software Engineering Plan• Testing/Validation procedures to be formalized


Role of SW MGR/COOD1. Prioritization and assignment of the SE Tasks2. Tracking activities across CCSM Groups (interface mgmt)3. Coordination (external outside CCSM)

To obtain shared codesTo track + help coordinate research codesTechnology tracking in generalAcross outside Collaborators (ACPI/NASA)

4. Software Decision Making within CCSM5. Release Coordinating/accountability/closure6. Represent SE interests/issues on SSC7. Single point of contact for other groups8. Maintaining SE plans9. SE training10 responsible for setting SE protocols


Action Items1. Erik write up CM procedure, Put in Developers Guide2. Internal meeting on building/testing. Bring desired features3. ACPI meeting. Extra day to iron out building/testing

Recommendations:1. SEI training for tech support2. Incrementally build up protocols and documents with goal of

better organizing the software process for ccsm.3. SE coordinator/manager added to present SSC


Functions Prioritization Closure Coordination

Scientific SE's working in "research mode"Collaborate with/ reviewed by Scientists Software/Utility SE's working on infrastructure developmentCoordinated by Software Professional

Negotiation of time between scientists and coordinatorCoord has power to make decisions about how things are done.

Coordination/Management


CCSM ORG ChartSSC

ATM OCN PALEO Software

scientist

programmer

research infrastructure

Coordinator


ORG Chart

research In fras tru c tu re

sc i p rog ram m er

S c ien tis t

A TM O S P H E R E O C E A N H A R D W A R E softw are

S S C coord in a to r

Move infrastructure work


THE PLAN: NCAR will organize the composition and management of its large modeling efforts to maximize the effectiveness of multidisciplinary teams.

THE PLAN: NCAR will ensure that the organizational structure of its major modeling projects integrates the expertise of both physical and computer scientists.

THE PLAN: NCAR and UCAR will develop a coordinated strategy for developing simulation models that allows the institution to be agile in allocating resources for the

development and evolution of all aspects of the end-to-end simulation environment.THE PLAN: Management of major software projects will

accommodate the broad constituency of the team effort, facilitated by project managers who have authority commensurate with their responsibility

NCAR Strategic Plan


THE PLAN: NCAR will adopt software engineering practices that promote high performance and the efficient development of large simulation models and software infrastructure.

• Establish formal software design procedures.

• Develop layered software

• Use software development tools.

• Object-oriented techniques and frameworks.

NCAR Strategic Plan


Computer = processor and memory layoutcomplexity = computation speed + data motionPerformance obtained by minimizing complexity for a particular

architecture and exploiting parallelism

Memory access is quantized in cache lines (4-16 words)Pay cache load cost whether use 1 value or all values in line

Registers fastest (on chip)L1 Cache (on chip, 2-3 clock cycles away)L2/L3 Cache (off chip)Local/remote memory

Scalar Processor Optimization


CCSM Issues

•Code Development Process (this meeting)• Who defines process?• Who coordinates the efforts?• What design docs are necessary?

• Integration with ACPI and NASA CAN efforts• Insufficient Software Education• Implementation of Software Engineering Plan•Testing/Validation procedures to be formalized


ACPI Update (rosinski)• Short duration DOE initiative to speed up climate models

• Oak ridge, LLNL, ANL, NASA DAO NCAR,

• 2-d decomp of ccm and modularize physics/dynamics• Initial focus on CPL and ATM (see drake web page)• Numerous targets completed (truesdale doing config manage)• 2-d specific problems: Transposition logic (FFT• Oak Ridge doing Eulerian spectral and Semi-lagrangian

spectral parallelization• LLNL (rotman/lin/mirin) doing Lin-Rood parallelization• 5.5 Years/day on 64 Winterhawk II processors standalone• Little impact from NetCDF output•CPL: two tracks: Dec freeze track w/ classic cpl

NGC track: In prototyping phase


NASA CAN (deluca) Increasing Interoperability and performance of Grand Challenge Applications in the earth, space, life and microgravity sciences

Testing/Validation crucial. Allows decision making and schedulingUp Front Design is very important.

Structure: things that people deal with day-to-dayProcess: How we do things. New stuff

Concentrate on structure things that help us do our job better.





Concentrate on structure things that help us do our job better.


Summary





Concentrate on structure things that help us do our job better.1st start with pieces, then bring it all together.


Misc items1. Configuration management2. Coding Convention3. Build procedures and user support4. Test scripts and validation5. Coupler interfaces


Supervision1. Many Ses continuing to report to scientists2. Some report directly to se coord/mgr3. coord/mgr reports to maurice/ssc