VDI 2.0 새로운물결의시작 CISCO SYSTEMS KOREA DATA CENTER SYSTEM ENGINEER SEUNG MAN YOO
VDI2.0 in Cisco DC 젂략 왜 GPU 인가? GPU 기반의데스크탑가상화기술소개 GPU 기반의시스코솔루션소개 Why Cisco
시스코데이터센터비젂과젂략 Defined by Applications. Driven by Policy. Delivered as a Service / Solution BUSINESS REQUIREMENTS Policy Policy Network Policy Compute Cloud BUSINESS OUTCOMES Business Agility New Business Models Lower TCO 3
시스코데이터센터비젂과젂략 Defined by Applications. Driven by Policy. Delivered as a Service / Solution Applications Security & L4-7 Services Management & Orchestration (M&O) Software Development (DevOps / PaaS) Indirect / Cross-Areas (Cisco & 3 rd party) Direct NETWORK COMPUTE CLOUD ACI UCS Private Cloud Stack SDN / Network Virtualization Integrated Stacks / Hyper Convergence Hybrid Cloud DC Switching / MDS Storage Intercloud 4
왜 GPU 인가?
VDI 1.1 멀티미디어에대한수요증가, VDI의역사 - VDI 2.0 을위한진화의시작 VDI BC RDP to W2K VM on ESX VDI.5 브로커기반의풀을통한프로비젂 Windws XP 값비싼서버와스토리지 VDI 1.0 씬프로비저닝, 동접데스크탑, 네할램 8GB dimms, E5 프로세서 16GB Dimms SSD 기반 Win7 VDI 2.0 여러대의모니터에대한필요, 애플리케이션성능증가 3D 엑셀레이션프로그램, 대부분 VDI 인프라의현위치 VDI Big Bang 물에서벗어나갓육지로나온물고기 사용자와애플리케이션의위치 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
2 개의프로세서로구성된윈도우 7/8 Co-Processors 참고해야할사항? CPU 1 to ~18 프로세싱코어 GPU 100 ~ 1000 개의병렬처리식프로세싱코어 소프트웨어개발업체는여러해동안애플리케이션이 GPU 를사용하도록코딩을해오고있음 GPU 는강화된애플리케이션 offload 를통해서 CPU 의성능과확장성을향상시킴 GPU 최적화애플리케이션은시스템램과 GPU 램을사용함 오피스 2013 은 GPU 를사용하지않을시에 20% 정도의 CPU 를더사용하게됨
계속해서늘어가는 GPU에대한요구사항 Today s Apps and OSs Demand GPUs
눈으로보는 GPU 효과 오피스 2013 에서는디폴트로그래픽엑셀레이션이활성화되어있음 GPU 동시프로세싱은스크린업데이트를위한 CPU 사용율을낮출수있음
눈으로보는 GPU 효과 하드웨어그래픽엑셀레이션비활성화시. X CPU 사용율이급격하게증가
눈으로보는 GPU 효과
VDI 유저프로파일별 GPU 요구사항 디자이너 Graphics and Media Professionals, Design Engineers 프로파일 에상관없이 CATIA, CS6, Inventor 파워유저 Financial Analysts, Traders, Design Reviewers PLM, Solidworks, Adobe Dreamweaver, Medical Imaging Showcase 일반오피스사용자 Office workers, productivity and line-of-business workers GPU 는모든유저에게보다웏등한체감성능제공 MS Office, Photoshop
GPU 기반의데스크탑가상화기술소개
GPU enabled VDI VIRTUAL MACHINE VIRTUAL DESKTOPS NVIDIA GRID Enabled Virtual Desktop Agent NVIDIA NVIDIA GRID ENABLED Hypervisor NVIDIA GRID GPU
가상데스크탑환경의 GPU 공유기술 GPU Pass-through 유저에게 1:1 로물리적으로 GPU 할당 Shared GPU 소프트웨어가상화기반 GPU 공유 Virtual GPU Nvidia Grid 소프트웨어기반하드웨어 GPU 가상화
Hardware Virtualization Software GPU Pass-Through App App App VDA App Guest OS Citrix XenDesktop 5.6 FP1 VMware view 5.2 Client Virtual Virtual Virtual Virtual VIRTUAL MACHINE NVIDIA Citrix Xenserver Vmware ESX Microsoft vcpu vmemory vstorage vnetwork GPU HYPERVISOR GRID K1/K2 UCS Server CPU Memory Storage Network GPU
Hardware Virtualization Software Shared GPU App App App VDA App App App Guest OS VDA App App App VDA App OS Virtual Virtual Guest OS Virtual Virtual VIRTUAL MACHINE Virtual Graphics vcpu vmemory vstorage vnetwork vgraphics Virtual Graphic s Virtual Graphic s Rendered Desktop Graphics APIs Client RemoteFX Microsoft Server 2008 R2 DX9 Microsoft Server 2012 DX9, 10, 11 VMware vsga - DX9, OGL2.1 Citrix Xenserver HYPERVISOR UCS Server NVIDIA GRID K1/K2 CPU Memory Storage Network GPU
Hardware Virtualization Software Virtual GPU App App App VDA App App App App VDA App App App App Guest OS VDA Virtual Virtual Guest OS Virtual Virtual VIRTUAL MACHINE VIRTUAL MACHINE NVIDIA NVIDIA NVIDIA Graphic Commands Standard Nvidia Citrix XenServer Vmware ESX Client Client Client vcpu vmemory vstorage vnetwork vgpu HYPERVISOR GRID Software Nvidia GRID VGX software UCS Server CPU Memory Storage Network GPU
Software sharing Virtual GPU Pass-through VIRTUAL MACHINE VIRTUAL MACHINE VIRTUAL MACHINE VIRTUAL MACHINE VIRTUAL MACHINE VIRTUAL MACHINE VIRTUAL MACHINE VIRTUAL MACHINE VIRTUAL MACHINE VIRTUAL MACHINE VIRTUAL MACHINE VIRTUAL MACHINE Operating System Operating System Operating System Operating System Operating System Operating System Operating System Operating System Operating System Operating System Operating System Operating System Applications Applications Applications Applications Applications Applications Applications Applications Applications Applications Applications Applications Soft Graphics Adapter Soft Graphics Adapter Soft Graphics Adapter Soft Graphics Adapter NVIDIA Graphics NVIDIA Graphics NVIDIA Graphics NVIDIA Graphics NVIDIA Graphics NVIDIA Graphics NVIDIA Graphics N NVIDIA Graphics N HYPERVISOR Translation Manager GRAPHIC COMMANDS HYPERVISOR NVIDIA vgpu Manager GRAPHIC COMMANDS HYPERVISOR NVIDIA Graphics vgpu vgpu vgpu vgpu vgpu vgpu vgpu DirectX and OpenGL Limitations
VDI 프로파일을위한 GPU 지웎요약 Vendor GPU Pass-Through GPU Soft- Sharing NVIDIA GRID Virtual GPU XenApp 6.5 on Windows Server (vdga) (vsga) (vsphere 6) (RemoteFX)
GPU 사이징 Nvidia GRID 카드 일반적용도가상데스크톱 NVIDIA GRID K1 vsga 최대 128 유저 vgpu 사용자별최적의 vgpu 선택 가상워크스테이션 NVIDIA GRID K1 or K2 vgpu 사용자별최적의 vgpu 선택 vdga full GPU power/ CUDA support @ 891 MHz @ 891 MHz @ 745 MHz @ 2,500 MHz
Cisco UCS C240 M4 GPU 가상화를위한최적의플랫폼 최대 2개의 Grid K1 이나 K2 카드장착지웎 인텔최싞하스웰기반의 2소켓 E5-2600 v3 CPU 장착 최대 768 GB 메모리용량제공 ( 32G 기준 ) 1, 10, 40 Gb/s 네트웍성능제공 CIMC 기반의단일서버관리소프트웨어제공 UCSM 기반의랙서버통합관리솔루션지웎
Cisco UCS C460 M4 GPU 가상화를위한최적의고성능플랫폼 최대 2개의 Grid K1 이나 K2 카드장착지웎 인텔최싞하스웰기반의 4소켓 E7 v3 CPU 장착 최대 6TB 메모리용량제공 ( 64G 기준 ) 1, 10, 40 Gb/s 네트웍성능제공 CIMC 기반의단일서버관리소프트웨어제공 UCSM 기반의랙서버통합관리솔루션지웎
! 1 2 3 4 Intel Inside XEON UCS C460 M4 Cisco UCS C240 & C460 M4 혁싞적인 FCoE 기반의랙서버통합솔루션제공 UCS C240M4 UCS C460M4 UCS 6200 FCoE 통합관리모듈을통한랙서버통합 UCS 6200 FCoE 통합관리모듈을통한랙서버통합 UCS C240M4 UCS C460M4
Cisco UCS B200 M4 GPU 가상화를위한최적의블레이드플랫폼 Mezz MXM(GRID) 타입의카드탑재 인텔최싞하스웰기반의 2소켓 E5-2600 v3 CPU 장착 최대 768 GB 메모리용량제공 ( 32G 기준 ) Mezz MXM(GRID) 타입의카드가 B200 M4에탑재 탑재모델은 NVIDIA TESLA M6 GRID 카드 성능은대략 NVIDIA GRID K2 카드와동일 릴리즈예정일은 CY2015 Q4( Roadmap )
Why Cisco
시스코데이터센터비젂과젂략고객사의서비스와인프라에최적화된통합솔루션벤더!! BUSINESS REQUIREMENTS Policy Policy Network Policy Compute Cloud BUSINESS OUTCOMES Business Agility New Business Models Lower TCO 27
THANK YOU