Big Data Analytics at Vestas Wind Systems

Title slide/end slide image dimensions:

Title slide images must always fit this size:

Height: 10.6 cm; Width: 25.4 cm

Title slide images must always follow this position:

Horizontal: 0 cm; Vertical: 0 cm

To change title slide/end slide image:

Contact Communication Support

Tip:

Vestas drawing guides will help you position and

crop images and objects. Images and objects will

snap to drawing guides when repositioned or

cropped.

To add drawing guides:

• Right-click on next to slide

• Select ‘Grid and Guides…’

• Check ‘Display drawing guides on screen’

• Press ‘OK’

Know the wind

Big data analytics at Vestas Wind Systems A/S

Anders Rhod Gregersen

[email protected]

Senior specialist, Vestas Wind Systems A/S

5th of June 2012, Anders Rhod Gregersen, Vestas Wind Systems A/S

mailto:[email protected]

Add presentation information via

View/Header and Footer

To add pre-formatted bullets please

use the increase/decrease indent

buttons found in the PowerPoint menu

Presentation Outline

• Introduction to Vestas

• Finding a good site and HPC

• Forecasting

• Usability

• Building HPC capability

• Road ahead

Vestas Wind Systems A/S 2

Introduction

What are we trying to achieve?

How do we do it?

Q&A







Marked leader in Wind turbines

Wind only company

Install base approx. 50GW / 50.000 wind turbines

22.000 employees

World wide






Why high performance data heavy computing?







Why HPC and Big data?

Renewable vs base production

• Predictability

• Integration

Significant investment

What matters to the customer?







The complexity of business case certainty


Site picture

Height contours

Complexity

Turbulence

Service Cost

Cost of Energy

Wind Resource






Finding a good site

Traditional process:

Point measurements (met mast)

Point estimate of wind resource

Point estimate of turbulence

Drawbacks:

Costly

Time consuming

Point measurement of flow

No weather context







Understand wind ressource

Hindcasting the weather

Very high area resolution

Very high time resolution

Unlike weather services, the model is sensitive to wind

Time series from turn of millenium onwards

Approx. 200 parameters

Wind measurements in context

Extreme events







Understand wind turbulence


Complex vs simple terrain

Turbulence and fattigue

Point measurement of turbulence

Modeling via Computational Fluid Dynamics

Moving from point to flow

Wind shear

In-flow angles






Flow over terrain







Architecture







Architecture overview

Compute:

1.306 compute nodes

2.612 Intel X5670 CPUs

15.672 X5670 cores

>34TiB RAM

>161TFLOPS

10.752 M2070Q cores

Interconnects:

4xQDR Infiniband (40Gb)

2:1 over-subscribed

Ethernet

Storage:

1.680 spindles

14 I/O servers

>20GB/s







Primary memory


Modeling fluid flow means

Solving Navier-Stokes equeations which means

Memory bound code

On die memory controller

One bank, three ranks/CPU

Remote memory via Message passing interface (MPI)

MPI via Infiniband

Low latency / high bandwidth






Secondary memory


GPFS – general parallel filesystem

Freedom in both bandwidth and capacity

Singular name space

I/O moves via Infiniband

Exposed to the OS as POSIX

Backup vs snapshots






Analytics at Vestas







Building database technology that scales beyond petabytes Use cases:

Point queries: user wants the full time series for the four adjacent points

Data exploration: meteorologist queries region or world for existence or frequency of phenomenon

Forensics: Engineer is doing a post-mortem on high frequency data from a wind turbine

Regular database/warehouse has problems scaling (conceptually/economically)







Business challenge

What we have

200 parameters

Hourly measurements

Time series since 2000

What we need:

Relational database functionality

Fast point queries

Scalable queries







Facts:

Nature of data:

Volume, Velocity

Assumption, all data time series

Assumption, common query full time series

Optimization: partition elimination

Optimzation: parameter elimination

Solution: a column based







Enter IBM InfoSphere BigInsights

Partnership with IBM

Joint development with

IBM Almaden (research) and

IBM Silicon Valley Lab (productization)

Weekly meetings







Advantages of a partnership: Software developed that matches our need

Software lifecycle in handled by IBM

Software productization


Title slide/end slide image dimensions:

Title slide images must always fit this size:

Height: 10.6 cm; Width: 25.4 cm

Title slide images must always follow this position:

Horizontal: 0 cm; Vertical: 0 cm

To change title slide/end slide image:

Contact Communication Support

Tip:

Vestas drawing guides will help you position and

crop images and objects. Images and objects will

snap to drawing guides when repositioned or

cropped.

To add drawing guides:

• Right-click on next to slide

• Select ‘Grid and Guides…’

• Check ‘Display drawing guides on screen’

• Press ‘OK’

Copyright Notice

The documents are created by Vestas Wind Systems A/S and contain copyrighted material, trademarks, and other proprietary information. All rights reserved. No part of the documents may be reproduced or copied in any form or by any

means - such as graphic, electronic, or mechanical, including photocopying, taping, or information storage and retrieval systems without the prior written permission of Vestas Wind Systems A/S. The use of these documents by you, or

anyone else authorized by you, is prohibited unless specifically permitted by Vestas Wind Systems A/S. You may not alter or remove any trademark, copyright or other notice from the documents. The documents are provided “as is” and

Vestas Wind Systems A/S shall not have any responsibility or liability whatsoever for the results of use of the documents by you.

Thank you for your attention






Usability challenge

HPC grew out of academia

What to expect from users?

Users range from PhDs

….to saleforce

One common tool for common users

….enables tracability

Point and click HPC for regular users

Terminal based interaction for the power user







Building up HPC capability

Q1/2003 First commercial license for CFD Q4/2006 First cluster (40 cores) Q3/2007 CFD model validated.

Q4/2008 Second cluster (15 TFLOPS, TiBs) Q2/2011 Third cluster (3rd largest industrial HPC, PiBs)
