Open Cloud Computing A Case for HPC - unizg.hr · 9 HPC Market Overview Server Revenue by IDC Competitive Segments Segment Price Range Supercomputer $500K and up $2.58 3.20% 1.50%

1

Open Cloud ComputingA Case for HPCCRO NGI DayZagreb, Oct, 26th

Philippe TrautmannHPC Business Development ManagerGlobal Education @ ResearchSun Microsystems, Inc.

2

Agenda

• The Cloud• HPC and Cloud: any needs?• Cloud Computing from Sun• Getting Started

3

The Cloud

3

● The illusion of infinite computing resources

● The elimination of an up-front commitment

● Pay for use of computing resources

4

What Does Cloud Mean? “ A fundamental shift in the computing paradigm”

- Steve Ballmer, CEO Microsoft“The return of the mainframe, and the mainframe is a set of computers. You never visit them, you never see them. But they're out there.”

- Eric Schmidt, CEO Google“It's nothing more than a faddish term for the established concept of computers linked by networks. A cloud is water vapour”

- Larry Ellison, CEO Oracle“You build your app, and you inherit our architecture”

- Marc Benioff, CEO SalesForce.com

“The Truth Is Rarely Pure And Never Simple”-Oscar Wilde

5

Public

Business ModelsPrivate Hybrid

You don’t know who else is on the

same server, network or disk

that you are

You own the server, network and disk, and decide who

gets to run on it with you

You own some parts and are sharing some

parts, though in a controlled way

6

Application Domains

Domains Drive Differencesin Hardware and Software Architecture

HPC

IntelligenceMedical

AnalyticsFinance

Web

7

Faster time-to-marketReduction of custom softwarePay only for what you useGrow infrastructure with business

Cloud Computing Layers

Applications offered on-demand overthe network (salesforce.com)

Basic storage and compute capabilities offered as a service (Amazon web services)

Developer platform with built-in services (Google App Engine)

Infrastructure as a Service

Platform as a Service

Software as a Service

8

8

HPC market requirements

8

9

HPC Market Overview

Server Revenue by IDC Competitive Segments

Segment Price RangeSupercomputer $500K and up $2.58 3.20% 1.50%Division $250K - $500K $1.30 1.60% -0.70%Department $100k - $250K $3.62 7.10% -0.04%Workgroup <100K $1.73 1.90% -0.06%

2009TAM $B

CAGR (07 – 13)

DOWN SIDECAGR

HPIBMDellOtherSun

HP

IBM

DELL

Other

SUN

IDC Server Revenue by Vendor 2008

IDC Estimates that for every $ spent on Servers● An additional $.39 is spent on storage● An additional $.25 is spent on services

IDC HPC Application/Industry ForecastServers Storage

Application Segment 2009 ($K) 2013 ($K) 2009 ($K)University Academic $1,800,235 $2,337,419 6.75% $571,344 16.27%Govt. Lab $1,425,431 $1,863,896 6.93% $433,087 12.33%Bio Sciences $1,217,297 $1,781,031 9.98% $652,271 18.57%CAE $952,761 $1,562,311 13.16% $455,087 12.96%Defense $871,585 $1,186,212 8.01% $414,288 11.80%EDA $613,729 $948,920 11.51% $173,687 4.94%DCC & Distribution $576,228 $835,046 9.72% $269,913 7.68%Geosciences & Geo Engineering $529,772 $807,039 11.10% $222,042 6.32%Weather $371,260 $545,329 10.09% $119,956 3.42%Economics /Financial $261,750 $421,115 12.62% $64,663 1.84%Chemical Engineering $223,468 $260,900 3.95% $88,262 2.51%Other $182,756 $140,644 -6.34% $20,227 0.58%Mechanical Design & Drafting $106,400 $98,205 -1.98% $27,568 0.78%Total Revenue $9,132,672 $12,788,067 4.10% $3,512,395 100.00%

CAGR (07-13) % of Mkt

10

The Importance of HPC

• Reduce costs and increase efficiency• Improve quality and be first to market• Make better and faster decisions• Applications becoming increasingly computationally intensive• Required to run more and more of these applications• Need to analyze more and more data

HPC can solve these problems and is now a required technology to stay competitive

Organizations are Under Pressure

11

• The “P” in HPC• Technical limitations – system, storage, interconnect,

complexity• Exploding Data Requirements

• Increasing fidelity of modeling and simulation• Instruments that spit out PetaBytes of Data• Requirement for collaborative research

• Complexity of Use• Need reliable solutions that are easy to architect, deploy and

use• Space, power and cooling issues

Limiting factors

12

Time toLoad Data

Time to Compute

Time toStoreData

2009 2011

You can only compute as fast as you can move the data

ExponentialData Growth

Time toLoad Data

Time toStoreData

Timeto

Compute

Barriers to HPC: Data Access

13

Barriers to HPC:I/O Bottlenecks – Application Enemy #1• Prevents applications from scaling• Leads to poor overall application performance• Complex – CPU? Memory? Storage?

Interconnect? Application?• Removing I/O Bottlenecks requires an end-to-

end approach

• Is there a catch for Cloud computing here?

14

IDC: Cloud costs vs.DataCenter

14

£0.00

£5,000,000.00

£10,000,000.00

£15,000,000.00

£20,000,000.00

£25,000,000.00

£30,000,000.00

Startup

cost

Year1

Year2

Year3

Year4

Year5

Year6

Year7

Year8

Year9

Year10

Data CentreCloud

In H2 2009, IDC analyzed the costs of running 100% of a typicallarge businesses IT infrastructure in a DC versus the cloud:

After year 3,cloud costs exceeded the DC

Final ScoreDC: £15MCloud: £26M

Even with 3 yearrefresh cycles of 30%,DC remains much cheaper

15

The Real Problems On The Horizon That Will Make Or Break Cloud Services

Standards Common standards on development, deployment, and

migration/transition Ability for businesses to move a system from one cloud to another

– No Lock In

Differentiation Different providers with different value propositions – Big vendors,

Telcos, Hosting Providers…. Specialization and the emergence of best of breed providers in

specific areas

Competition Price competition and eco system competition No one dominant winner

16

16

Cloud Computing From Sun

16

Cloud Architecture

17

Sun’s StrategyDevelop the core technologies for Sun's Open Cloud Platform

Offer Services through Sun's public cloud service – the Sun Cloud

Work with service providers andenterprises to build their own clouds

Develop open standards

Build partnerships and communities

18

Cloud Architecture – Future

Partner and Build

User Apps and ServicesInternet Accessible APIs and UIs

Servers Storage Network

Virtualized Datacenter Management Layer

Customer Web Site

StorageService

QueuingService

JavaEEService etc.

Application Catalog,Forums, DocsVirtual Datacenter

Management Console

Accounting, Billing and Metering

Identity Service

DatabaseService

ComputeService

19

Building Robust Sun Cloud Ecosystem

Sun Cloud

20

Initial Public Cloud Roadmap

Internal AlphaStorage

Compute

Early AccessStorage

Compute

Update 1Storage, Compute

Adds Identity, Queuing, Database services

Q1 2009

H2 2009

Q2 2009Sun Open Cloud

PlatformWork with Customers on

Product Version of Software in Public Cloud

21

21

Getting Started

21

22

Adopt Models & Standards

=

23

Focu

s H

ere

Faster, Better, Cheaper, Reduced Scope, or Someone Else's Problem?

Prepare your IT Service Management

24

Consider Adoption StrategiesTest and

Development

Functional Offload (Batch Processes –

TimesMachine)

Functional Offload (Storage – SmugMug)

Augmentation(Temporary Load – Animoto)

Web Service

25

Profile Applications & Workloads

Suitable for cloud• Time based• Very parallel (i.e. batch)• Spiky traffic• Capital intensive

(especially startup)• Proof of Concept• Low utilization• Less deployment costs• High bandwidth costs /

high real estate

Not suitable for cloud• Vertically scaled applications• Consistent load levels• Latency sensitive

applications• Insecure applications• Hardware device dependent

(e.g. fax server, SNA gateway)

• ISV unsupported• Per CPU licensed

applications

26

• Participate in the Development of our Open Cloud APIs

• Sign up for Early Access to Sun Cloud Services

• Become a Sun Cloud Partner• Let Sun experts help you take

advantage of Cloud Computing

http://sun.com/cloud

Getting Started Today

27

philippe.trautmann@sun.com

sun.com/hpc

THANK YOU!