High Performance Computers - QUDIT

31
5/5/2011 1 High Performance Computers and how they can improve your research? and how they can improve your research? CRICOS No. 000213J a university for the world real R Mr Mark Barry [email protected] 3/05/2011 Overview: HPC and Research Support In brief: We provide support to QUT’s researchers (incl. P/G students) in their use of High Performance Computing (HPC), advanced visualisation, data analysis tools and … Original group was established in 1992 – twentieth anniversary next year! CRICOS No. 000213J a university for the world real R anniversary next year!

Transcript of High Performance Computers - QUDIT

5/5/2011

1

High Performance Computers

and how they can improve your research?and how they can improve your research?

CRICOS No. 000213Ja university for the worldrealR

Mr Mark [email protected]

3/05/2011

Overview:HPC and Research Support

In brief:

We provide support to QUT’s researchers (incl. P/G students) in their use of High Performance Computing (HPC), advanced visualisation, data analysis tools and …

Original group was established in 1992 – twentieth anniversary next year!

CRICOS No. 000213Ja university for the worldrealR

anniversary next year!

5/5/2011

2

Beyond the desktop…

Our main areas of Focus:

– High performance computational and graphics computing systems– A wide range of State of the art software to support research– Client support consultancy and training services– Research collaboration technologies and other e-Research systems

http://www.itservices.qut.edu.au/hpc/

CRICOS No. 000213Ja university for the worldrealR

Extending to new points on the Research Continuum…

HPC group assists with ways of doing Research which go beyond traditional modes of enquiry:traditional modes of enquiry:

– Traditional Modes – Theoretical &/or Experimental– 3rd Mode – Simulation– 4th Mode – Large Scale Data Exploration/ Analysis, incl. visualisation etc

e-Research = research using non-traditional modesi i th t bl d E i t Th Si l ti ICT &/ D t

CRICOS No. 000213Ja university for the worldrealR

i.e. enquiry that blends Experiment, Theory, Simulation, ICTs &/or Data

5/5/2011

3

e-Research & the ‘3rd/4th Modes’ - a HPC perspective

Computational simulation &/or large scale data analysis:

• are fundamentally both Theoretical & Experimental• are fundamentally both Theoretical & Experimental

• frequently need a multi-disciplinary approach across traditional boundaries

• may involve dispersed teams, remote resources

CRICOS No. 000213Ja university for the worldrealR

• typically generate/manipulate huge data sets

• requires all things HPC; data access, transfer, mining, analysis and visualisation tools; collaboration platforms; etc

e-Research & the ‘3rd /4th Modes’ @ QUT

QUT’s HPC facilities strike a balance between:

• Capability for “heroic simulations”• Capacity for “production” simulations

that add to steady stream of progress

Potential Benefits of HPC/E-Res.adoption include increased:

• Productivity

CRICOS No. 000213Ja university for the worldrealR

• Productivity• Innovation• Research competitiveness

5/5/2011

4

Putting the pieces together

Aspects of e-Research such as advanced ICTs, HPC, collaboration tools, large scale data analysis, mining etc are becoming increasingly important in many areas of scholarly inquiry @ QUT.

You may need re-invent the way you think and do research…

CRICOS No. 000213Ja university for the worldrealR

How can we help in your research?

Through:

• Provision of HPC infrastructure• Computational science support services• Domain expertise

CRICOS No. 000213Ja university for the worldrealR

Along with highly qualified staff who understand the research process...

5/5/2011

5

HPC systems and research software: how they can help?

CRICOS No. 000213Ja university for the worldrealR

Our local compute facilitiesTwo HPC systems

• 96 processor SGI Altix 4700 (Vega)96 processor SGI Altix 4700 (Vega)– Single SMP Machine– 198 GB RAM

• 400 processor SGI Altix XE Cluster (Lyra) – 41 node configuration

(2 Quad-Core processors, 16Gb RAM) (2 Si C 24/48GB RAM)

CRICOS No. 000213Ja university for the worldrealR

(2 Six-Core processors, 24/48GB RAM)– 960 GB RAM Total– 2011 upgrade: refresh of system

components and additional compute/memory

5/5/2011

6

Our local data facilities• High performance file service, hpc-fs

– 4.6TB Fibre Channel disk,– 40TB SATA Disk,,– 120TB Tape storage (2011 x 4)

• Very High performance scratch disk

– 24TB Panasas disk.

• Data can be accessed either from HPC compute resources or your desktop.

CRICOS No. 000213Ja university for the worldrealR

The management of research data is gaining attention within QUT – HPC is currently involved in projects to gather knowledge on what infrastructure and services are required by the research community to manage their data.

HPC System Configuration

Lyravega

File server

ssh/sftp/Xwindows

Network drive

lyra

CRICOS No. 000213Ja university for the worldrealR

hpc-fs

5/5/2011

7

Statewide & National compute facilities

• QUT is a member of QCIF*– QCIF has a SGI Altix cluster

• QCIF (ergo QUT) participating in numerous Federally funded projects including:

• RDSI

• NECTAR

• NCI

• …

CRICOS No. 000213Ja university for the worldrealR

Our Local HPC Lab• Local facility providing access to higher-end workstations,

typically memory and graphics.• Targeted for postgraduate students and researchers

• Range of Workstations– Dual Processor, 64bit Workstations.– Range memory configurations (4GB, 8GB, 16GB, 24GB) – Multi-screen setups, including 30”LCD monitors.– 3D Visualisation equipment – stereo projectors.– 3-D Scanners (laser and contact)

CRICOS No. 000213Ja university for the worldrealR

5/5/2011

8

Virtual HPC Lab

• Remote access to high-end workstation capabilities.• Targeted for postgraduate students and researchers

• Targeted towards researchers that require more than their desktop/laptop but the traditional HPC system does not meet there requirements – typically software not supported on Linux or is more of an ‘interactive’ user.

• Pool of ‘virtual machines’, which are connected to through a remote client.

CRICOS No. 000213Ja university for the worldrealR

Software Available

MATLAB (and SYMULINK)R

ABAQUSANSYS MSC.NASTRAN MSC.PATRAN

NAG Libraries Intel Compilers and toolsGraphics Libraries

Mathematical Modelling Environments

Finite Element Analysis PackagesSoftware Development Tools

MiscellaneousData Analysis Tools

Data Visualisation

MayaRapidform

ANSYS CFXFLUENT

3D Modelling EnvironmentsCFD Packages

MSC.Marc LS-DYNA

AVS/Express Volume Visualisation tools VTK & ITK

CRICOS No. 000213Ja university for the worldrealR

Gaussian09SPSS SAS NUD*IST Nvivo

5/5/2011

9

Scientific Libraries and Programming Support

• programming language support lib i f i tifi ti d• libraries of scientific routines and applications (Intel MKL, NAG, etc)

• general programming advice • code optimisation• parallelisation

CRICOS No. 000213Ja university for the worldrealR

p

What this means for you…

• Near limitless computational “grunt” available to QUT researchersavailable to QUT researchers

• Major research software packages and programming paradigms available

• Dedicated support specialists on-hand to help guide research and/or solve problems

CRICOS No. 000213Ja university for the worldrealR

help guide research and/or solve problems

5/5/2011

10

In other words…

Fast computer systems+ leading-edge programs + dedicated research support staff

= Time savings !!

CRICOS No. 000213Ja university for the worldrealR

Qualitative Data Analysis

CRICOS No. 000213Ja university for the worldrealR

5/5/2011

11

Qualitative ResearchMy General rule of thumb:

•If you can count it   > Quantitative

•If you can’t count it  > Qualitative

• see through the eyes of the research subject, narrative teller, historical subjects etc.,

• to understand the descriptive contexts 

Qualitative research tends to be based on an interpretative perspective that seeks to: 

and processes that underpins your 

observations (holistic perspective),

Qualitative MethodsOther Ways of collecting Data

• Participant Observation– It requires that the 

• Unstructured Interviewing– maybe some initial guiding q

researcher become a participant in the culture or context being observed.

• Direct Observation– a direct observer doesn't 

typically try to become a participant in the context

• Focus Groups/Structured

questions or core concepts to ask about, but there is no formal structured instrument or protocol

• Case Studies– an intensive study of a specific 

individual or specific context. A combination of methods (e.g., 

• Focus Groups/Structured Interviews

– Structured questions and informal discussion

unstructured interviewing, direct observation) can be used.

5/5/2011

12

Mixed Methods Research

• Combine Quantitative and Qualitative methods in the one studythe one study

• Do Qualitative and Quantitative work that follow on from each other as part of the same research project

Qualitative Data AnalysisQualitative Seminars/Workshops

Introduction to Qualitative Methods

Introduction to Nvivo

Advanced Nvivo

Introduction to Leximancer

Advanced Leximancer

5/5/2011

13

Consultations

• Free consulting service to assist you/direct you in gaining the skills in qualitative data analysis

Qualitative Data Analysis

• Students have found advice useful in association with advice from supervisors, school/faculty research consultants and peers.

• Please note: these sessions are aimed at getting you up and running in relevant qualitative procedures within the tools -- we won’t do your analyses for you!

• If you need directions/help with qualitative analysis, research data design, ethnography, interviews, or other Qualitative methods or geospatial social data feel free to make contact

CRICOS No. 000213Ja university for the worldrealR

data feel free to make contact.

Quantitative Data Analysis

CRICOS No. 000213Ja university for the worldrealR

5/5/2011

14

Main focus• Using machine learning algorithms/data mining procedures for

predictive data model building

Quantitative Data Analysis

predictive data model building • The analysis of quantitative data using machine learning algorithms

- Classification and Regression Trees, Tree Ensemble methods, Regression Splines and Genetic Algorithms

Secondary focus• Common statistical procedures and the application of SPSS to

data analysis• Survey Methods

CRICOS No. 000213Ja university for the worldrealR

Survey Methods

Online Survey Tools:K SKey Survey

Queensland University of Technology

CRICOS No. 000213J

5/5/2011

15

Key Survey

• Key Survey is an online survey creation tool.

• HPC provides this as a free service for all staff and students (conditions apply).

• Key Survey is widely used with over 2200 existing surveys and 126 000 respondents accommodated over the past year.

CRICOS No. 000213Ja university for the worldrealR

year.

Some Key Benefits

• Easy & Dynamic Use –• ‘Wizard-driven’ interface

• Adaptable and FlexibleAdaptable and Flexible

• Flexible distribution methods• Email, URL, Embedding, Hard Copy, Web Page Pop-up

• Customisation –• Customisable Question Types and Logic

• QUT and associated branding

• Templates and Plug-ins

• In-Line HTML, Javascript and CSS

• Intelligent Reporting –

CRICOS No. 000213Ja university for the worldrealR

• Customisable and informative reports

• Export to multiple formats for better delivery.

5/5/2011

16

2011 News

• New Key Survey website launched

• New QUT visual templates

CRICOS No. 000213Ja university for the worldrealR

Case Studies

• Code porting/optimisation• Scientific visualisation

CRICOS No. 000213Ja university for the worldrealR

5/5/2011

17

Code Porting and Optimisation

CRICOS No. 000213Ja university for the worldrealR

Mark [email protected]

Code Porting and Optimisation

Simulation code can take a long time to complete

Need to speed code up

Translate computational code to a 'faster' language

Profile code

Variable declarations and usage

Optimise memory usage/addressing

Optimise looping strategy and branching statements

CRICOS No. 000213Ja university for the worldrealR

branching statements

Parallelise code

5/5/2011

18

CRICOS No. 000213Ja university for the worldrealR

Code Porting and Optimisation

Unsure of the specifics of the code (for the most part, it's arbitrary)

Example 2 - Parameter Sweep (200000s ~ 55hours)

Port code from Matlab to C++

Data structures specific for new generation Intel processors

Loop restructuring and indexing

Intel compilers

Final Time: 1.53 seconds

CRICOS No. 000213Ja university for the worldrealR

5/5/2011

19

Code Porting and OptimisationExample 3: Processing Extreme Scale Datasets

Some datasets are getting really, really big ...

Lidar Data: 200 Million Point CloudIntersection with aerial photography

I t ti Vi li ti f T b t Different Compression

Algorithms

Reorder the internal storage data structures for cache coherency

Specialist low memory implementation of statistical

Interactive Visualisation of Terabytedatasets using commodity hardware

implementation of statistical algorithms and methods

Web interfaces to access Terabyte data (Lidar, WorldScope)

Massive Scale Visualisation

Specialist FPGA / GPU Compute Facilities to Accelerate Scientific Applications

CRICOS No. 000213Ja university for the worldrealR

5/5/2011

20

FPGA – field programmable gate arraysHighly specialised computing devices for:

• Reconfigurable computing

• Digital design

• H/W algorithm implementation

• Signal / image / data processing

• Cryptographic applications etc

CRICOS No. 000213Ja university for the worldrealR

GPU – General Processing UnitWhat is GPU Computing?

CRICOS No. 000213Ja university for the worldrealR

5/5/2011

21

GPU: Matlab pluginsSpeedup using “A = GPUsingle(rand(n,n))”

CRICOS No. 000213Ja university for the worldrealR

Scientific Visualisation

CRICOS No. 000213Ja university for the worldrealR

5/5/2011

22

Volume VisualisationBone fracture healing

CRICOS No. 000213Ja university for the worldrealR

Volume VisualisationBone fracture healing

CRICOS No. 000213Ja university for the worldrealR

5/5/2011

23

Volume VisualisationBone fracture healing

CRICOS No. 000213Ja university for the worldrealR

Volume VisualisationBone fracture healing

CRICOS No. 000213Ja university for the worldrealR

5/5/2011

24

Point Cloud VisualisationLaser + Receiver System = Lidar Data

(Light Detection and Ranging)

CRICOS No. 000213Ja university for the worldrealR

Visualisation of Lidar DataClassification of terrain and vegetation

CRICOS No. 000213Ja university for the worldrealR

5/5/2011

25

Visualisation of Lidar DataPoint by point overlay of Google Map Data

CRICOS No. 000213Ja university for the worldrealR

Flow VisualisationFlow VisualisationCharacterisation of flow in gross Characterisation of flow in gross

pollutant trappollutant trap

CRICOS No. 000213Ja university for the worldrealR

5/5/2011

26

Application of modern image-based flow visualisation techniques

highly detailed flow images

CRICOS No. 000213Ja university for the worldrealR

GPU-Based Visualisation

• flow animations– direction + orientation

• 40,000 velocity points per frame

• GPU – advecting flow and blending frames

• virtual dye experiments

CRICOS No. 000213Ja university for the worldrealR

5/5/2011

27

GPU-Based Visualisation

CRICOS No. 000213Ja university for the worldrealR

Visualisation Summary

• HPC provides advanced visualisation services to assist in your research

• Images and animations that can be used for papers and presentations– relay information in a visually interesting and easy-to-understand wayy y g y y

• Data processing and visual exploration of data– can reveal patterns and details in the data that may not otherwise be obvious

• Development of custom tools to allow interaction with visualisations

• Services that may be beyond the capabilities of researchers

CRICOS No. 000213Ja university for the worldrealR

5/5/2011

28

Research / Scientific Data Issues

CRICOS No. 000213Ja university for the worldrealR

Data – The Big Picture

• Advances in digital sensors, communications, computation, and storage have created huge collections of data, capturing information of value to business, science, government, and society., , g , y

• There was an estimated 281 exabytes (2.25 × 10^21 bits) of data in 2007.(data: information created, captured, or replicated in digital form)

• More digital data is being created than there is storage to host it.- 2007 marked the crossover year in which more digital data was created than there was data storage to host it.

CRICOS No. 000213Ja university for the worldrealR

• With the "Data Deluge", it is becoming increasingly important for data to be well managed so that it can be retrieved when needed and be easily discovered by others if that data is in the public domain.

2008 International Data Corporation (IDC) white paper

5/5/2011

29

What is Research Data?

• Research Data is any digitally encoded information generated while engaging in research.

• Research data can be text files, spreadsheets, presentation slides, images, video, computational simulation results, DNA sequences, sensor readings, emails, receipts and much more.

CRICOS No. 000213Ja university for the worldrealR

Questions to Ask Yourself• Do you have some idea of the data that you will generate as part of your

studies/research? What type of data? How much?

• Do you have somewhere safe to store your data?y y

• Do you have a plan for making backups of your data.

• Do you have 'file naming conventions' that you will use to identify file content.

• Do you have a system for documenting how you generated/obtained your data?

• Is your data sensitive or confidential in nature such that you need to make sure that access to the data is restricted?

CRICOS No. 000213Ja university for the worldrealR

sure that access to the data is restricted?

• Will you be able to find your data when it comes time to writing your thesis?

• Who will own your data at the completion of your studies/research?

5/5/2011

30

HPC Data Management Services

• HPC File Share

• Researcher Data Management Websitehttp://researcherdata.qut.edu.au

• Mediaflux Data Repository

A t St t d N ti l t i f t t

CRICOS No. 000213Ja university for the worldrealR

• Access to State and National storage infrastructure

In Summary...

Queensland University of Technology

CRICOS No. 000213J

5/5/2011

31

QUT’s HPC and Research Support group

We provide local support to QUT’s researchers in their use of High Performance Computing (HPC) advanced visualisationHigh Performance Computing (HPC), advanced visualisation, data analysis tools …

Provide a bridge for accessing wide range of additional services external to QUT which compliment and supplement our services.

CRICOS No. 000213Ja university for the worldrealR

??CRICOS No. 000213Ja university for the worldreal

R