+ All Categories
Home > Technology > HPC To Every Desk

HPC To Every Desk

Date post: 05-Dec-2014
Category:
Upload: eerola
View: 494 times
Download: 1 times
Share this document with a friend
Description:
Techila brings HPC to every desk. It is the missing link between applications and the computing capacity available in the Clouds.Techila gives unlimited and easy scaling of MATLAB performance in the Cloud, linear acceleration of stochastic R scripts, and easy integration to any project with Open APIs.
37
May 10, 2012 Tuomas Eerola HPC To Every Desk
Transcript
Page 1: HPC To Every Desk

May 10, 2012

Tuomas Eerola

HPC To Every Desk

Page 2: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

Outline

Introduction Demo Proof-of-Concept and Use-Cases Benefits Evaluating

Page 3: HPC To Every Desk

INTRODUCTION

Page 4: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

“…to win the game we need to have the superb tools and processes.”

http://www.youtube.com/watch?feature=player_detailpage&v=L1C3heqnclk#t=158s

Markus Ekman CIO, SVP of Information Management, VTT

Page 5: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

Techila Technologies A Finnish university spin-off. Founded to solve a genuine customer

problem: Computing power exists, but applications cannot utilize it.

Page 6: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

Techila is the missing link between

applications and the computing

capacity available in the Clouds.

Page 7: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

Missing Link?

Computing Capacity

Need for Computing Capacity

PrivateCloud/Intranet

Page 8: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

The Missing LinkNeed for Computing Capacity

Computing Capacity

PrivateCloud/Intranet

MISSING LINK

Page 9: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

What Does My End-User Get?Speed-up also to applications, which have not been able

to benefit of conventional server-based computing solutions.

Unlimited and easy scaling of MATLAB performance in the Cloud, without license limitations.

Linear acceleration of stochastic R scripts, without overloading the System Administrator issues. Enabled by

autonomic computing.

Easy integration to any project with Open APIs.

Page 10: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

What Does My IT Manager Get?Complementing solution to cluster schedulers, like the

GridEngine, SLURM, Windows HPC Server. Offer secure high throughput and high performance computing services to new

user groups.

Versatile reporting and billing.

Harness and manage capacity from various Clouds (public, private, hybrid…). Enable seamless cloud-bursting onto

leading cloud services. Shielding to Users from complexity.

Easy system administration, enabled by autonomic computing.

Page 11: HPC To Every Desk

DEMO

Page 12: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

Can I See It In Practice?

http://www.youtube.com/watch?v=VwbD2jc0qA0

Page 13: HPC To Every Desk

PROOF-OF-CONCEPT

Page 14: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

Who Are The Users? Perfect for optimization, modeling,

simulations or data-analysis. Perfect for projects who aim on

minimum Time-To-Publication. Reference cases from Finance,

Biosciences and Pharma, Space, Oil and Gas, Civil Engineering,…

Irreplaceable for users who need faster access to results in their business-critical computing.

Page 15: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

From The Users”With the computing power of Windows Azure, the over 21-hour simulations can now be done in less than 10 minutes……We are excited to see that the Windows Azure with Techila integration allows using the simulation model interactively and getting instant feedback for changes in model assumptions.”

Päivi Ojala, Chief Actuary, Fennia Life

”The fact that computational resources provided by the cloud service are available, increases prospects to conduct really demanding projects that were not possible few years ago.”

Sampsa Hautaniemi, Adjunct Professor, Bioledicum

Page 16: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

CASE: 3D Animation Animation rendering happens a

frame at a time. Techila middleware can provide

near-linear acceleration to 3D renderer.

Blender 3D animation renderer plug-in was cloud-enabled using the Techila middleware.

Benefits: No impact on user experience. Results come in a fraction of

time.16

Click to watch on YouTube:http://www.youtube.com/watch?v=O7JxSt6X3Mo

Page 17: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

CASE: Machine Learning RF-ACE is an efficient

implementation of a robust machine learning algorithm for uncovering multivariate associations.

Exploring associations is a CPU intensive but embarassingly parallel computation.

Techila middleware distributed the workload across a collection of 1000 CPUs, cutting down computation from years to days.

17

Page 18: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

CASE: Surface science (1/2) Surface science to determine geometrical and electronic structure of

surface. Obtaining a comprehensive picture of the unconventional

superconductors relies very much on electronic spectroscopes. Need to run computations with multiple variables, check the results with

experimental predictions and adjust the combinations of parameters and run again.

High-Performance Computers (HPC) with Message Passing Interface (MPI) was tested at NERSC

Challenges: Scalability issues Frustration resulting of HPC queuing and MPI’s complexity

Need for a simple, scalable and fast solution

18

Page 19: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

CASE: Surface science (2/2) Originally a FORTRAN code. Code was changed from

hardcoded to accept command line parameters.

Currently using MATLAB as front-end; Preprocessing, Postprocessing

Benefits: Ease of input and output handling Modularity Easy visualization

19

Page 20: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

CASE: Bridge simulation Mitigation of Stay-Cable Vibration. Reduce wind hazards on long-span bridges. MATLAB application using Monte carlo methods

20

Page 21: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

CASE: 3D Inversion Geophysics inverse problem. Hundreds of measurement points. Variable frequency. Distribution brings great benefits.

21

Page 22: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

Improved a deterministic non-distributable problem through stochstic approach and clever use of large pool of underutilized on-premise IT-infra

Approach relies on the large number of grid nodes rather than on the actual raw computational power

More nodes better guesses faster optimization. Estimates can now be found significantly faster by using

Techila. 2 days vs. 2 months Techila benefits:

Time limits (stop after time) Optimization (fastest resources) Failure tolerance (long project) Ease of use (agility)

CASE: Optimization

22

Page 23: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

CASE: Enhanced p-value accuracy Statistics/ informatics. Code written in R language. Using LME4 package, which is a

package to fit linear and generalized linear mixed-effects models.

Indepdendent sImulations. The more simulations, the more accurate p-value.

Distributed the code using the Techila with R language integration.

Techila took care of autonomous LME4 distribution to Workers.

Techila provides a linear speed-up.23

Page 24: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

CASE: Medical imaging Inverse research applications

on medical imaging. Creating imaging algorithms

for EIT to build accurate, cheap and even portable devices that can save people’s lives.

Dentists always aim at implants that stay safely in place. The screw needed for that should be drilled in a deep as possible.

24

Page 25: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

CASE: Image reconstruction Scientists have had to accept

inaccurate imaging in their research.

Image enhancement applies stochastic models, which are computing-intensive

Stochastic image enhancement has not been feasible. Takes 1 month.

Techila supports scientific innovation.

Enables working on newest data as medical image

reconstruction can be done in 5 minutes

25

Images: Uygar Tuna, Sari Peltonen, Ulla Ruotsalainen. Department of Signal Processing, Tampere University of Technology. Gap-Filling for the High Resolution PET Sinograms with a Dedicated DCT Domain Filter. 2009. Data acquired by the ECAT High Resolution Research Tomograph (HRRT, CTI PET Systems, Knoxville, TN, USA), located at Turku PET Centre.

Page 26: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

CASE: Data analysis Systems Biology Pattern Discovery. Calculating statistics for

SNP (single-nucleotide polymorphism) pairs.

Relatively data-intensive. Still, 8955% speed-up with

the idle capacity of ~100 computers.

26

Page 27: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

CASE: Combinatory analysis Assessing the likelihood that a primary

breast cancer tumour develops metastases.

A lot of genes and their combinations that may contribute to the metastasis progression.

Key challenge on finding the right combination of genes from among 25000 genes.

Windows Azure with Techila integration used.

1200 Windows Azure instances running MATLAB code.

15-year project completed in 4,5 days. 27

Page 28: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

CASE: Model calibration Monte Carlo simulation has become an essential tool in

financial engineering GARCH models and other volatility models Exotic options LIBOR models Risk measures

Drawback: No closed-form analytical solution for the price

Price available only through Monte Carlo simulation.

Makes the model calibration computationally expensive Calculating primes of the each subinterval separately,

and then collectig the results enables distributed computing.

Near-linear speed-up 28

Page 29: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

HPC To Every Desk

Innovation + Efficiency

= Competitiveness

Enable high-performance computing also in

applications, which have not been able to benefit of

conventional, server-based solutions.

Page 30: HPC To Every Desk

BENEFITS

Page 31: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

Powering The Existing Applications

Techila enables applications to harness and

utilize the infinite capacity of the Cloud

quickly and easily.

Page 32: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

No License Limitations

Unlimited scaling of MATLAB performance in the Cloud, without license limitations.

Stochastic simulation, data-analysis, optimization, modeling,…

+ 1 MATLAB and compiler license

+ Techila middleware

+ Windows Azure

= Unlimited scalability.

Page 33: HPC To Every Desk

CONFIDENTIAL

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

Ease Of Use And Security

Unlike any other HPC middleware solutions, Techila can be integrated

easily and used

securely, without the complexity.

Page 34: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

Deployment in Minutes

Deployment automated with Techila Deployment Tool: Just type in your Windows

Azure Subscription ID and select region.

No installation required on End-User computers: Extract Techila SDK and link to

your Azure address and Certificate.

Page 35: HPC To Every Desk

EVALUATING

Page 36: HPC To Every Desk

The content of this presentation is confidential. Using, copying, or giving any part of this presentation or its description to a third party is strictly forbidden without the prior written consent of Techila Technologies Ltd.

How Do I Get Techila?Pay-as-you-Use or Permanent (Perpetual) license.

Techila Technologies and Microsoft promotional offer to Universities and Academia: Free Techila to

users, who sign up for Azure.

Designed to enable easy evaluation of benefits.

Limited-time offer is valid until end of June 2012.

Please contact Tuomas Eerola/ Techila Technologies for full offer details.

Page 37: HPC To Every Desk

TUOMAS EEROLAVICE PRESIDENT, PARTNERTECHILA TECHNOLOGIES LTDe: [email protected]

t: +358 50 336 7730


Recommended