GPU Technology Conference 2016 S6194: Delivering Graphics Intensive - - PowerPoint PPT Presentation

gpu technology conference 2016 s6194 delivering graphics
SMART_READER_LITE
LIVE PREVIEW

GPU Technology Conference 2016 S6194: Delivering Graphics Intensive - - PowerPoint PPT Presentation

GPU Technology Conference 2016 S6194: Delivering Graphics Intensive Applications to Computing Labs and BYOD in Education Michael Goay, Executive Director of IT USC Viterbi School of Engineering mgoay@usc.edu November 6, 2015 Board of Councilors


slide-1
SLIDE 1

November 6, 2015 Board of Councilors

GPU Technology Conference 2016 S6194: Delivering Graphics Intensive Applications to Computing Labs and BYOD in Education

Michael Goay,

Executive Director of IT USC Viterbi School of Engineering mgoay@usc.edu

slide-2
SLIDE 2

Session Objectives

  • Graphics‐enabled VDI @ USC Viterbi
  • Options to deliver graphics‐rich applications in:

– physical computing lab – virtual computing lab

  • Graphics‐enabled VDI implementation

– common pitfalls to avoid

slide-3
SLIDE 3

Agenda

  • USC Viterbi School of Engineering
  • Challenges
  • Solution
  • Benefits
  • Next steps
  • Takeaways
slide-4
SLIDE 4

USC Viterbi School of Engineering

  • Founded 1905
  • 8 Academic Departments:

AME, ASTE, BME, CEE, CHEM/MS, CS, EE, ISE

  • Academic Programs:

27 BS, 58 MS, and 13 PhD

  • Students: 2700 UGs, 5300 Gs
  • Faculty: 185 TT and 110 NTT
slide-5
SLIDE 5

USC Viterbi School of Engineering

  • ~5% annual growth in MS

students over past 5 years

  • 30 computer classrooms
  • 900+ desktops/laptops
  • 200+ instructional software
  • 1300+ class sections per

semesters

  • 18 FTE’s, 50 student workers
slide-6
SLIDE 6

Challenges

  • Digital natives; mobile
  • More classes are teaching with graphics intensive

applications

  • Applications require ever‐increasing processing

power to run with acceptable performance

  • Increase in class enrollment
  • Costly to build and maintain

classrooms with computers

slide-7
SLIDE 7

Project Goals

Empowering students to study with cutting‐edge performance, graphics‐rich applications anywhere, anytime, on any device Enhancing mobility to improve student collaboration, productivity and innovation

slide-8
SLIDE 8

Project Goals

Gaining flexible, rapid application deployment capabilities to enhance teaching and learning with latest engineering and design applications Empowering faculty to experiment with innovative teaching methods

slide-9
SLIDE 9

Viterbi Virtual Desktops Relieve Space Constraints

Virtual Desktops From Any Device From Any Location

Auditoria General Classrooms Outdoor Classrooms

slide-10
SLIDE 10

Solution

vGPU‐enabled Virtual Desktop Infrastructure

slide-11
SLIDE 11

Defining Terms

  • VDI – Virtual Desktop Infrastructure
  • Horizon – VMware Horizon; VDI product
  • Virtual Desktop – virtual computer hosted on a

server

  • Host – server cluster that hosts the desktops
  • Client – end‐point device
  • Connection Broker – service that connects a client

to a virtual desktop

slide-12
SLIDE 12

Instructional Computing

  • In the business of application delivery
  • Install : Local execution; on‐premises access
  • Stream : Local execution; on‐premises access;

flexible

  • Present : Server execution;
  • n‐premises + remote access; flexible
slide-13
SLIDE 13

Application Delivery ‐ Install

  • Characteristics:
  • Local program execution
  • On‐premises access
  • Pros:
  • Use local hardware resource
  • Cons:
  • Inflexible
  • Risk of software conflicts

Solutions: Sneakernet, sftp, download manager, Microsoft SCCM, Dell KACE, LANDesk, Symantec Altiris

slide-14
SLIDE 14

Application Delivery ‐ Stream

  • Characteristics:
  • Local program execution
  • On‐premises access
  • Pros:
  • Use local hardware resource
  • Flexible computing
  • Cons:
  • Limited to LAN

Solutions: Citrix Provisioning Server (PVS), Dell vWorkspace (formerly WSM)

slide-15
SLIDE 15

Application Delivery ‐ Present

  • Characteristics:
  • Programs execute on server
  • Remote access
  • Pros:
  • Flexible computing
  • Anywhere, anytime, any device
  • Cons:
  • High upfront cost; limited graphics support

Solutions: Microsoft RDS, Dell vWorkspace, VMware Horizon, Citrix XenDesktop

slide-16
SLIDE 16

App Delivery ‐ Comparisons

Flexible Computing Performance Operational Risk Impact on Learning Experience

Install

Low High –h/w dependent Low Medium – compatibility issues

Stream

High High – h/w dependent Medium – High Medium –

  • nsite only

Present

High Medium – High NVIDIA GRID the enabler High High – anywhere, anytime, any device; a clear winner!

slide-17
SLIDE 17

When it Happened

  • Desktop streaming in physical labs:

2008 Q3 ‐ Present

  • VDI Proof of Concept : 2013 Q4 – 2014 Q2
  • VDI Pilot : 2014 Q3 – 2015 Q2 (Academic Year

2014‐2015)

  • VDI Phase I : 2015 Q3 – Present (Academic Year

2015‐2016)

slide-18
SLIDE 18

Workstation Solution Architecture – VMware

Shared GPU – vSGA (Horizon View 5.2) Direct GPU – vDGA (Horizon View 5.3) Hardware Virtualized GPU (Horizon 6)

Up to DirectX 9.0c and OpenGL 2.1 only; no workstation‐class graphics support Aka GPU Pass‐Through DirectX 11 and OpenGL 4.3; workstation‐class graphics Flexible workstation‐class graphics support with high application compatibility

Guest OS

virtual gfx driver

Guest OS

virtual gfx driver NV driver ESX w/ vSGA ESX w/ vDGA

Guest OS

NV driver

2 GPUs/GRID K2 4 GPUs/GRID K1

Guest OS

NV driver

2 GPUs/GRID K2 4 GPUs/GRID K1

ESX Hypervisor

vGPU vGPU 2 GPUs/GRID K2 4 GPUs/GRID K1

slide-19
SLIDE 19

NVIDIA vGPU Profiles (K2/K1)

GRID K1 and GRID K2 : each GPU has 4GB of frame buffer memory Tesla M60 and M6 (Aug 2015) : each GPU has 8GB of frame buffer memory

slide-20
SLIDE 20

GRID K2 vGPU Profiles @ USC

Each Dell PowerEdge R730 or R720 server has 2x GRID K2 cards.

Virtual GPU Frame Buffer (MB) vGPUs/ GPU vGPUs/ Board vGPUs/ Server Applications

GRID K260Q 2048 2 4 8 ArcGIS, Citilabs Cube, Siemens NX, SolidWorks, STAR‐CCM+, etc. GRID K240Q 1024 4 8 16 Adobe Creative Cloud suite: Photoshop, Dreamweaver, Illustrator, etc. GRID K220Q 512 8 16 32 Autodesk AutoCAD, COMSOL, Erwin, Mathematica, MATLAB, Minitab, MS Office, Visual Studio, Python, Questa Sim, R, Revit, SAS, Xilinx ISE, etc.

slide-21
SLIDE 21

VM Configurations

  • High Graphics Intensity Apps
  • Supported apps: ArcGIS, Autodesk Maya, Citilabs Cube, Siemens NX,

Siemens Solid Edge, SolidWorks, STAR‐CCM+, etc.

  • 4 vCPUs, 32GB RAM, K260Q vGPU profile (2GB vRAM), 2 displays
  • Medium Graphics Intensity Apps
  • Supported apps: Adobe Creative Cloud suite – Photoshop, Dreamweaver,

Illustrator, etc.

  • 2 vCPUs, 16GB RAM, K240Q vGPU profile (1GB vRAM), 2 displays
  • Light Graphics Intensity Apps
  • Supported apps: Autodesk AutoCAD, COMSOL, Microsoft Visio, Microsoft

Office, Mathematica, MATLAB, Minitab, R, SAS, Xilink, etc.

  • 2 vCPUs, 8GB RAM, K220Q vGPU profile (512MB vRAM), 2 displays
slide-22
SLIDE 22

Tech Specs Horizon 6 Architecture

slide-23
SLIDE 23

Technical Specs ‐ Hardware

  • Dell SonicWall SuperMassive firewalls
  • Dell Force10 (10GbE) network switches
  • Dell EqualLogic iSCSI SAN storage arrays
  • Dell PowerEdge R730/R720 servers
  • NVIDIA GRID K2 cards
  • Teradici APEX 2800 PCoIP offload cards
slide-24
SLIDE 24

Technical Specs – Software

  • vSphere ‐ Cloud computing

virtualization system includes ESXi (hypervisor)

  • Horizon ‐ Desktop virtualization system
  • UX ‐ User experience monitoring
  • PernixData FVP ‐ software‐based

storage acceleration solution

  • TrendMicro Deep Security anti‐malware
slide-25
SLIDE 25

USC‐Specific User Experience

  • Simplified identity management – students

use same USC NetID account login credentials to access the Viterbi VDI system

  • Granular user entitlement management –

users are authorized/prioritized access to VDI based on class enrollment, class schedule, and where the client is connecting from

  • Unified desktop experience of data storage

across physical and virtual computing environments

slide-26
SLIDE 26

VDI Benefits to Students

  • Increases mobility
  • Enables device independence
  • Gains flexible computing
  • Realizes enhanced performance, productivity, and

collaboration

  • Improves security of user data via regular central

data backup

slide-27
SLIDE 27

VDI Benefits to Faculty

  • Affords the same flexibility and mobility
  • Focuses on academics, not technology
  • Enables classrooms to transform to customizable

learning spaces

  • Empowers faculty to experiment with innovative

teaching methods

slide-28
SLIDE 28

VDI Benefits to Administration

  • Relieves space constraints; eases classroom

scheduling

  • Facilitates online classes to include graphics‐

intensive applications

  • Facilitates innovative teaching methodologies
slide-29
SLIDE 29

VDI Benefits to IT

  • Eases system maintenance
  • Offers more secure computing environment
  • Affords agile and flexible computing
  • Affords rapid deployment of applications
  • Reduces break/fix issues at users’ devices
  • Focuses on more strategic projects
slide-30
SLIDE 30

Next Steps

  • Force multiplier… expand VDI capacity; capitalize on

new gen GPU, Tesla M60 cards and GRID 2.0

  • Convert workstations in computer classrooms to

thin clients

  • Optimize VDI management
  • Enhance disaster recovery capabilities
  • Faculty/staff desktop replacement
slide-31
SLIDE 31

Takeaways

  • Focus on the user experience
  • Know your users and use cases
  • Understand software licensing / behavior
  • Get the right people involved
  • Conduct a pilot that is true to scale
  • Optimize at every level and scale for growth
  • Prepare for change
slide-32
SLIDE 32

Final Thoughts

  • VDI gives flexibility, but implementation is complex
  • Virtualization has needs and considerations
  • New pressure on infrastructure
  • Changed cost structures ‐ cost shifted from edge to

data center

slide-33
SLIDE 33

Your Opportunity

Great rewards often require bold risk – so step up to face your giants!

GRID Test Drive http://www.nvidia.com/object/vmware‐trygrid.html NVIDIA and VMware Community (NVC) http://www.nvidia.com/object/nvidia‐vmware‐community.html

slide-34
SLIDE 34

Questions and Comments