PowerPoint Presentation

Size: px

Start display at page:

Download "PowerPoint Presentation"

미려 진
6 years ago
Views:

1 CAE on AWS Jungin Lee Oct 2016

2 Agenda 1 WHAT IS AWS? 2 WHY CAE on AWS? 3 HOW ANSYS on AWS? 4 CASE STUDIES

3 한국을포함, 총 13 개의도시에 AWS Region 이위치하며, 각 Region 은 2 개이상의 Availability Zone 의독립적데이터센터로이루어져있음

APIs On demand Uniform Compute Scaling Security CDN Backup

Networking Messaging Pay as you go Available

4 AWS 위에서는서버확장 / 축소가불과몇분안에일어나며, API 를통한자동화를이용하여비즈니스의민첨함과순간확장성을극대화함 Auto Scaling Case Study Rich APIs On demand Uniform Compute Scaling Security CDN Backup DNS Database Storage Load Balancing Workflow Monitoring Networking Messaging Pay as you go Available ec2-run-instances ami-b232d0db --instance-count 5 --availability-zone eu-west-1c --instance-type m1.medium 4

5 현재약 70 여개의서비스를제공중 TECHNICAL & BUSINESS SUPPORT Support Professional Services Partner Ecosystem Training & Certification Solutions Architects HYBRID ARCHITECTURE Integrated Networking Direct Connect Identity Federation Integrated App Deployments Business Apps ANALYTICS Data Warehousing Business Intelligence Hadoop/ Spark Streaming Data Analysis Streaming Data Collection Machine Learning Elastic Search Identity Management Business Intelligence APP SERVICES Queuing & Notifications Workflow Search Transcoding Access Control DevOps Tools Key Management & Storage MOBILE SERVICES API Gateway Identity Sync Mobile Analytics Single Integrated Console Push Notifications MARKETPLACE Security SECURITY & COMPLIANCE Monitoring & Logs DEVELOPMENT & OPERATIONS One-click App Deployment DevOps Resource Management Application Lifecycle Management Containers Triggers Resource Templates Configuration Compliance Networking Web application firewall Databases IoT Rules Engine Device Shadows Device SDKs Device Gateway Registry Assessment and reporting Storage ENTERPRISE APPS Virtual Desktops Sharing & Collaboration Corporate Backup Resource & Usage Auditing Account Management Data Backups Compute VMs, Auto-scaling, & Load Balancing Storage Object, Blocks, Archival, Import/Export CORE SERVICES CDN Databases Relational, NoSQL, Caching, Migration Networking VPC, DX, DNS Security & Pricing Reports Integrated Resource Management Regions INFRASTRUCTURE Availability Zones Points of Presence

6 AWS Korean Customers

7 Agenda 1 WHAT IS AWS? 2 WHY CAE on AWS? 3 HOW ANSYS on AWS? 4 CASE STUDIES

Examples 고에너지물리학 기상모델링 유체, 구조, 재료분석 유전, 단백질및분자역학 반도체및유전시뮬레이션 3D

8 Scalability for a Simulation-Driven World Discovery, analysis, visualization 점차중요해지는 Simulation-Driven, Data-Driven 디자인및개발 Examples 고에너지물리학 기상모델링 유체, 구조, 재료분석 유전, 단백질및분자역학 반도체및유전시뮬레이션 3D 렌더링및시각화 리스크분석 딥러닝 Cloud unlocks data-driven analysis at massive scale

9 사이클의더많은반복은, 비용효율적인생산으로.. Engineering Design (CAD) Conceptual Design Data & Process (PLM) Simulation & Analysis (CAE) Tooling Design (CAM) Production

10 Engineering / Design Simulation (CAE) FEA Thermal CFD Crash Electro Magnetic

11 CAE 와같은 HPC 는클라우드가제공하는가치를최적으로활용할수있으며, AWS 는가장많은경험과파트너를축적하고있음 Time 결과도출시간절감 Cost 비용절감 Collaboration AWS 글로벌인프라활용

12 CAE 와같은 HPC 는클라우드가제공하는가치를최적으로활용할수있으며, AWS 는가장많은경험과파트너를축적하고있음 Time 결과도출시간절감 Cost 비용절감 Collaboration AWS 글로벌인프라활용

13 CAE 효율성 = 컴퓨팅자원의유연성 Predicted Demand Actual demand What size of cluster do you need? 디자인과해석워크로드에따라사실은다른종류와규모의컴퓨팅이필요함 결국 CAE 에서의시간의혁신은 HPC 컴퓨팅자원의한계를어떻게뛰어넘는가의문제로귀결 And what kind of cluster is it? Large Memory? More and faster Cores? Faster Storage? Faster Networks? What Generation of Processor?

14 HPC 클러스터의유지관리는 HPC H/W 는막대한 Long-Term 투자? 언제가가장완벽한구매 Timing 인지? 얼마나구매해야하는지? 어떤스펙으로구매해야하는지?. 과연이러한부분이조직의 Core Competency 인가?

15 매우큰컴퓨팅파워를요구하는, 유체역학시뮬레이션 Job 이하나있다고가정하면..

16 대부분현재부서에서보유하고있는클러스터의사이즈가너무작아오래걸린다고하면..

17 회사차원에서 HPC 클러스터 Farm 을구축하여 Job 을처리하려고합니다.

18 하지만, 이역시 Job Queue 가생기게되고기다리는 Job 이생기게됩니다.? 다시처음으로반복됩니다..

19 Job Queue 의폐해 HPC 사용자는가장빠르게결과를얻고싶어하며, 기존의한정된자원을점유하기위해내부적으로 Queue 를만들게됨 IT 관리자는이미구매한리소스에대해최대한의 Utilization 을원함 Fewer simulations = lost opportunity! 자연스레 Job Queue 는 IT 자원을관리하기위한 Buffer 의역할을하게되어 Delay 가일어나게됨

20 반면, 클라우드에서는내가원할때마다동적으로 1-Click 으로원하는사이즈의 HPC Farm 을구성합니다.

21 Small Job 은그에맞게

22 그보다조금더큰작업도필요에만큼

23 Job Queue 없이병렬적으로!! 누구도기다리지않습니다.

24 Job 의종류에따른다양한인스턴스타입 Scale up and scale out

25 정리하자면 Job A = 1 core x 500 hours

26 정리하자면 Job A = 1 core x 500 hours Job A = 500 core x 1 hour

27 사례 ) 시간에따른서버 Core 수

28 CAE 와같은 HPC 는클라우드가제공하는가치를최적으로활용할수있으며, AWS 는가장많은경험과파트너를축적하고있음 Time 결과도출시간절감 Cost 비용절감 Collaboration AWS 글로벌인프라활용

AWS EC2 Consumption Models 온디맨드요금제 시간당요금제로내가원하는시간에만과금 Pay for

workloads, or to define needs 예약요금제 1 년혹은 3 년을예약하여 30~60% 할인 Make a

hourly charge 스팟요금제 AWS Region 의유휴자원에대한입찰 온디맨드가격의약 20% Bid for

29 AWS EC2 Consumption Models 온디맨드요금제 시간당요금제로내가원하는시간에만과금 Pay for compute capacity by the hour with no long-term commitments For spiky workloads, or to define needs 예약요금제 1 년혹은 3 년을예약하여 30~60% 할인 Make a low, one-time payment and receive a significant discount on the hourly charge 스팟요금제 AWS Region 의유휴자원에대한입찰 온디맨드가격의약 20% Bid for unused capacity, charged at a Spot Price which fluctuates based on supply and demand

30 On-demand 가격알아보기 66% Saving! 500 Cores 365 일사용 (100% Utilization) 500 Core 한달 20 일사용 (66% Utilization) 500 Cores 하루 8 시간사용 (33% Utilization) $17K/mo $11K/mo $5K/mo

31 AWS Spot Market 100% Spot On On-demand Reserved capacity 0% Capacity Over Time

32 Best Practices for Using Spot Stateless Fault tolerance Multi-AZ Loosely coupled Instance Flexibility HPC 와완벽한 Fit 을가진요금제

33 HPC Users Love Spot

34 Bid Price Vs Market Price You pay the market price

35 Bid Price Vs Market Price 75% Bid 50% Bid 25% Bid You pay the market price

36 Spot 가격알아보기 500 Cores 365 일사용 (100% Utilization) 500 Core 한달 20 일사용 (66% Utilization) 500 Cores 하루 8 시간사용 (33% Utilization) $17K/mo $11K/mo $5K/mo $2~3K/mo $1~2K/mo $0.5~1K/mo 80~90% More Saving!!!

37 CAE 와같은 HPC 는클라우드가제공하는가치를최적으로활용할수있으며, AWS 는가장많은경험과파트너를축적하고있음 Time 결과도출시간절감 Cost 비용절감 Collaboration AWS 글로벌인프라활용

38 AWS 를통한 GE 의제조협업플랫폼운영 (CEED) Cloud provides a global, distributed, secure, and scalable environment for collaborative design and manufacturing

39 서울포함, 전세계 13 개의지역의 Region 보유

40 Collaboration is More Secure in the Cloud Bring the users to the data, don t send the data to the users

41 Collaboration is More Secure in the Cloud Bring the users to the data, don t send the data to the users

P2, G2 Instance Types P2 8 NVIDIA Tesla K80 Accelerators ( 각각 2 개의 NVIDIA GK210 GPU) 을지원 각 GPU 는 12GB 메모리를가지고있으며 ( 초당 240GB 메모리대역폭제공 ) 2,496 개의병렬코어를지원 최대 20Gbps 대역폭지원 G2 Intended for remote graphics

42 P2, G2 Instance Types P2 8 NVIDIA Tesla K80 Accelerators ( 각각 2 개의 NVIDIA GK210 GPU) 을지원 각 GPU 는 12GB 메모리를가지고있으며 ( 초당 240GB 메모리대역폭제공 ) 2,496 개의병렬코어를지원 최대 20Gbps 대역폭지원 G2 Intended for remote graphics rendering and streaming High-performance NVIDIA GPUs, each with 1,536 CUDA cores and 4GB of video memory Each GPU features an on-board hardware video encoder designed to support up to eight real-time HD video streams

43 Application Streaming Middleware

44 Thin Client Remote Collaboration

45 Agenda 1 WHAT IS AWS? 2 WHY CAE on AWS? 3 HOW ANSYS on AWS? 4 CASE STUDIES

46 Scaling Out Simulation: Platform Strategy Simulation Trends Desktop Platform Multiphysics Simulation Systems Engineering Robust Design Simulation Democratization Enterprise Platform ANSYS EKM HPC Data Mgmt. Process Mgmt.

47 ANSYS Enterprise Cloud

48 ANSYS Enterprise Cloud: HPC on AWS Auto-scaling HPC provisions resources on-demand, using machine configurations optimized for specific workloads. Scale on demand 수요에따른서버확장 Match compute instances to workloads 워크로드에따른서버타입매치 Optimize steady state (reserved) and ondemand AWS spend

49 각고객마다분리되고고립된 Private 환경에서구동됨 (AWS VPC)

50 Solution Deployment ANSYS Enterprise 는고객마다별도로분리된 Account 에배포됨 ( 논리적망분리및고립 ) CloudFormation 템플릿을통해몇번의클릭만으로환경세팅이가능함 이로인해 AWS 의 Region 을다양하게확장하여사용가능 Chef 를통해탄력적인인프라배포 전체 Deployment 가 3 시간안에가능함 ( 기존에수개월이걸렸던것과비교 )

51 Agenda 1 WHAT IS AWS? 2 WHY CAE on AWS? 3 HOW ANSYS on AWS? 4 CASE STUDIES

협업을위한데이터저장, 공유, 배포 제조수율분석을위한 Big Data Running drive-head simulations at

52 Rapid Innovation in Electronic Products Hitachi 의자회사였으나 Western Digital 로인수됨 HGST applications for engineering: 분자동역학시뮬레이션, CAD, CFD, EDA 협업을위한데이터저장, 공유, 배포 제조수율분석을위한 Big Data Running drive-head simulations at scale: 수백만의병력적인파라미터를변경하며수개월이걸리는시뮬레이션을단몇시간안에완료함 Peak 시최대 85,000 Core 를사용함 (Spot Instance)

53 Scale Drives Faster Innovation at Honda Cloud offers us an opportunity, as we can innovate faster than before. - Ayumi Tada, IT System Administrator, Honda R&D Scalable Materials Simulations Before: 80 HPC nodes, 1 year to complete all needed simulations After: Scalable, on-demand HPC cluster on AWS Up to 1000 nodes, 16,000 cores at peak More simulations means more accurate results at lower cost

54 ResAssure 1M simulation in 1 day on AWS cloud Stochastic Simulation 은아직도 대다수의유전저류엔지니어 가수년씩걸리는 시뮬레이션을하루만에 가능하게합니다.. Dr Andrew Wadsley Chief Technology Office The company demonstrated ResAssure's capability, by carrying out 1 million realizations in 1 day for a real field Cartesian model of: 200,000 grid cells against 28 sensitivity parameters using 40 standard processing cores on AWS cloud Source:

55 자전거디자인 CFD 분석 Trek applications for engineering: CFD Simulations for bicycle design: 병렬적으로많은시뮬레이션을구동함 자전거시장에서매우중요한드래프팅기술을좌우하는디자인요소에대한통찰력제공

Azure Stack – What’s Next in Microsoft Cloud

Azure Stack – What’s Next in Microsoft Cloud Microsoft Azure Stack Dell EMC 와함께하는하이브리드클라우드전략 Microsoft Korea, Cloud+Enterprise 사업부진찬욱부장 Sr. Product Marketing Manager, Azure Stack Azure Momentum 120,000 New Azure customer subscriptions/month 715 Million