Nutanix AHV with NVIDIA Virtual GPU Solutions Ready to Meet the - - PowerPoint PPT Presentation

▶
nutanix ahv with nvidia virtual gpu solutions
SMART_READER_LITE
LIVE PREVIEW

Nutanix AHV with NVIDIA Virtual GPU Solutions Ready to Meet the - - PowerPoint PPT Presentation

Jul 06, 2023 162 likes •590 views

Nutanix AHV with NVIDIA Virtual GPU Solutions Ready to Meet the Demands of Any Workload Malcolm Crossley , AHV GPU Architect Tanuja Ingale , AHV Product Manager GTC, Mar 2019 Nutanix: The Enterprise Cloud Company Our Mission: Nutanix makes IT

slide-1
SLIDE 1

Nutanix AHV with NVIDIA Virtual GPU Solutions

Ready to Meet the Demands of Any Workload

Malcolm Crossley, AHV GPU Architect Tanuja Ingale, AHV Product Manager

GTC, Mar 2019

slide-2
SLIDE 2

2

Nutanix: The Enterprise Cloud Company

Our Mission: Nutanix makes IT infrastructure invisible with an enterprise cloud platform that delivers the agility and

economics of the public cloud, without sacrificing the security and control of on-premises infrastructure.

11,490+ customers

Over 145+ countries

6 continents

NTNX on Nasdaq

Founded in 2009 4,380+ employees

slide-3
SLIDE 3

3

Infrastructure

Time consuming to provision Multiple points

  • f failure

Process

Difficult to scale and upgrade Large upfront CapEx

People

Little time for innovation Requires IT specialists

IT Complexity is Hurting Business

slide-4
SLIDE 4

4

The Need for Desktop Virtualization (VDI)

Reduce cost and complexity of app and desktop management Simplify support and enable choice of BYO devices Increase employee productivity with anywhere access Centralized security to protect sensitive information Comprehensive desktop virtualization for any use case

slide-5
SLIDE 5

5

VDI and GPU workload evolution

High Density High Performance

VDI, Client/Server & business apps

  • Nutanix Comprehensive desktop virtualization
  • 100s of general knowledge & task workers per

Compute + Expansion module

  • ~2000 VDI users per X rack units w/o GPU

acceleration

  • 200+ VDI sessions with accelerated virtualized

graphics Complex Design & Visualization

  • High-end design, CAD, Rendering, Ray tracing

(M&E, Manufacturing)

  • Support 10s of specialized users (Design) per

Compute + Expansion module

  • Support AI, Deep ML, Compute workloads for

Improved VR perf

  • HPC workloads, Monte Carlo analysis
slide-6
SLIDE 6

6

GPU Infrastructure

Operational & workflow complexity Silos for graphics & compute Demand for GPU accelerated workloads

Process

Need predictable scaling VDI CapEx Underutilization Underutilization of True pre- emptive capabilities of GPU

People

Requires experts for GPU workflows/sizing

How IT complexity translates to GPU environments

slide-7
SLIDE 7

7

Innovating in Three Fronts for VDI/GPU accelerated env

Architecture

Scale out, software- defined architecture

Infra CapEx Preservation

Purchase based on user needs & provide 100% utilization

Technology

Powerful data fabric and control fabric capabilities

slide-8
SLIDE 8

8

Innovating in Three Fronts for VDI/GPU accelerated env

Architecture

Scale out, software- defined architecture

Infra CapEx Preservation

Purchase based on user needs & provide 100% utilization

Technology

Powerful data fabric and control fabric capabilities

slide-9
SLIDE 9

9

VDI Pain points on legacy architecture

Centralized Storage (SAN/NAS) Storage Network

  • Complex to manage
  • Costly to scale
  • Performance bottleneck
  • Difficult to scale
  • Managed separately

Servers

  • Commoditized by multicore

processors and dense memory

  • Managed by Hypervisor
slide-10
SLIDE 10

10

Virtualization App App Integrated, scale-out compute and storage Virtualization App App

Storage Controller Storage Controller Storage Controller Storage Controller

Server Server

Storage Controller Storage Controller

Built-in virtualization and management

Nutanix: Web-Scale Converged Architecture

slide-11
SLIDE 11

11

Built for Virtualization

▪ Software-defined approach with Controller VM per node ▪ Pooled storage resources across the platform and scale as needed ✓ High performance ✓ Auto optimized Nutanix Controller VM

(one per node)

slide-12
SLIDE 12

12

The Result is Linearly Scaling

VMs (Desktops) Number of Nodes

  • Scale incrementally one node at a time
  • Protect infrastructure investment by eliminating forklift upgrades
  • Scale storage capacity & performance linearly

Pay-as-you-grow

slide-13
SLIDE 13

13

Distributed Storage Fabric App Mobility Fabric Acropolis Prism AHV (built-in Virtualization)

Nutanix Platform

Infrastructure / VM Management Operational Insights Capacity Planning

Comprehensive management solution radically simplifying datacenter operations Turnkey infrastructure platform that converges compute, storage and virtualization to run any application, at any scale

slide-14
SLIDE 14

14

Nutanix AHV

The hypervisor built for the Enterprise Cloud

slide-15
SLIDE 15

15

AHV: Foundation of the Enterprise Cloud OS

+ + =

Amazing Management Enterprise Feature Set Lean Virtualization Cloud Operating System

NX Appliance OEM 3rd Party

slide-16
SLIDE 16

16

AHV Powers HCI

Security

  • Data Encryption
  • Certifications

Management

  • Prism Central
  • API and CLI

Self-Healing

  • Hot Spot Remediation
  • Auto STIG Compliance

Automation

  • Distributed Scheduler
  • Calm Blueprints
  • Biz. Continuity
  • HA
  • Synch Rep (2019)

Analytics

  • Prism Pro
  • Performance

Auditability

  • Logging
  • Remote Syslog

Extendable

  • Memory
  • vCPU

VDI/ GPU Acceleration

  • GPU, vGPU
  • PVS, MCS

Performance

  • AHV Turbo
  • IO Optimization

Ecosystem

  • Over 100 ISV solutions
  • Backup, SAP HANA
slide-17
SLIDE 17

17

Starting with Citrix XD/XA 7.9 … AHV is integrated with MCS

AHV: Citrix App and Desktop Virtualization

Citrix Infrastructure

VM VM VM VM VM VM VM VM VM VM

End user desktops

  • XenApp & XenDesktop
  • NetScaler VPX
  • ShareFile
  • XenApp, XenDesktop support
  • Cloud Connect InstantOn
  • PVS and MCS plug-ins
  • NVIDIA vGPU support
  • High perf. + data locality

Nutanix AHV is

Nutanix AHV

slide-18
SLIDE 18

18

Nutanix-NVIDIA Strategic Partnership

  • Nvidia: Industry leader in Visual Computing Technologies

and GPU accelerators

  • AHV
  • First commercial kernel-based virtual machine (KVM)

to support vGPU

  • Fully supports NVIDIA virtual GPU technology (GRID)
  • Quadro Virtual Data Center Workstation (vDWS)
  • nVIDIA GRID Virtual PC
  • GRID Virtual Applications
slide-19
SLIDE 19

19

AHV: Crystal Clear Graphics

GPU

vGPU

Tesla M10. Tesla M60 Tesla P4 Tesla P40. Tesla v100 Tesla T4

Multi- vGPU VM vGPU Live Migration vGPU HA

(Coming Soon) (Coming Soon)

slide-20
SLIDE 20

20

AHV: Modes of GPU usage

AHV Hypervisor

GPU driver Apps Guest OS VM

GRID GPU

Passthrough

GRID GPU vGPU vGPU vGPU vGPU

GPU driver Apps Guest OS VM GPU driver Apps Guest OS VM GPU driver Apps Guest OS VM

NVIDIA GRID vGPU Manager

slide-21
SLIDE 21

21

Acropolis GPU resource management concepts

Physical GPU 87:00.0 Vendor: Nvidia Type: PT-Compute Device: M60 Windows VM B GPU config: Vendor:Nvidia Type: PT-Compute Device:M60

NVIDIA Tesla M60

Physical GPU 86:00.0 Vendor: Nvidia Type: PT-Graphics Device: M60 Physical GPU 07:00.0 Vendor: Nvidia Type: PT-Graphics Device: M10

NVIDIA Tesla M10

Physical GPU 06:00.0 Vendor: Nvidia Type: PT-Graphics Device: M10 Physical GPU 08:00.0 Vendor: Nvidia Type: PT-Graphics Device: M10 Physical GPU 09:00.0 Vendor: Nvidia Type: PT-Graphics Device ID: M10 Linux VM C GPU config: 2 x Vendor:Nvidia Type: PT-Graphics Device:M10 XenApp VM A GPU config: Vendor: Nvidia Type: PT-Graphics Device:M60 Linux VM D GPU config: 2 x Vendor: Nvidia Type: PT-Graphics Device:M60

Not enough GPU resource

slide-22
SLIDE 22

22

Virtual GPU 86:00.0 Vendor: Nvidia Type: Virtual Device: M60-1Q Index: 1 Virtual GPU 86:00.0 Vendor: Nvidia Type: Virtual Device: M60-1Q Index: 3 Virtual GPU 86:00.0 Vendor: Nvidia Type: Virtual Device: M60-1Q Index: 2

Acropolis GPU resource management concepts

Physical GPU 86:00.0 Vendor: Nvidia Type: PT-Compute Device: M60 Xendesktop VM C GPU config: Vendor: Nvidia Type: Virtual Device ID: M60-1Q

NVIDIA Tesla M60

Physical GPU 87:00.0 Vendor: Nvidia Type: PT-Graphics Device: M60

XenApp VM D GPU config: Vendor: Nvidia Type: PT-Graphics Device: M60

Xendesktop VM A GPU config: Vendor: Nvidia Type: Virtual Device: M60-2Q Virtual GPU 87:00.0 Vendor: Nvidia Type: Virtual Device: M60-1Q Index: 0 Xendesktop VM B GPU config: Vendor: Nvidia Type: Virtual Device: M60-2Q Virtual GPU 86:00.0 Vendor: Nvidia Type: Virtual Device: M60-1Q Index: 1 Virtual GPU 87:00.0 Vendor: Nvidia Type: Virtual Device: M60-2Q Index: 0

Not enough GPU resource

slide-23
SLIDE 23

23

AHV VM GPU resource configuration

slide-24
SLIDE 24

24

Prism UI - Physical GPU overview

slide-25
SLIDE 25

25

Prism UI - Physical GPU metrics

slide-26
SLIDE 26

26

Prism UI - Virtual GPU metrics

slide-27
SLIDE 27

27

Prism UI – Multi-vGPU per VM

slide-28
SLIDE 28

28

Powerful REST API for all GPU resource information

slide-29
SLIDE 29

29

Acropolis 1-click operations and GPU resources

UVM GPU VM

AHV version 1 AHV version 2 AHV version 1

GPU VM

AHV version 2 AHV version 2 AHV version 1

GPU VM

slide-30
SLIDE 30

30

Innovating in Three Fronts for VDI/GPU accelerated env

Architecture

Scale out, software- defined architecture

Infra CapEx Preservation

Purchase based on user needs & provide 100% utilization

Technology

Powerful data fabric and control fabric capabilities

slide-31
SLIDE 31

31

Distributed caching of vDisksand VM data read by multiple CVMs

Nutanix Shadow Clones

~50% reduction in boot time

Nutanix Data & Control Fabric Solutions

Offloads virtualization tier → higher ops performance

Software-defined Snapshots & Clones

Array-based quick-clones for efficient provisioning Native VM-centric snapshots

0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 300 600 1200 1500 3000 Seconds Number of Virtual Desktops Relative Application Performance

Consistent response time while incrementally scaling blocks

slide-32
SLIDE 32

32

Nutanix Frame: Desktop-as-a-service

Browser based Desktop-as-a-service platform

Nutanix Frame

Multi-cloud Infrastructure (orchestrates workloads in public clouds or on-prem(AHV) Exceptional End-user & Admin UX

Xi Frame

slide-33
SLIDE 33

33

Innovating in Three Fronts for VDI/GPU accelerated env

Architecture

Scale out, software- defined architecture

Infra CapEx Preservation

Purchase based on user needs & provide 100% utilization

Technology

Powerful data fabric and control fabric capabilities

slide-34
SLIDE 34

34

Need for GPU powered & oversubscribed infra

  • Increased need for GPUs
  • AI Composability & Virtualization
  • Data Simulation
  • 20x Faster > CPUs
  • Fast Complex Calculations & Simulations
  • Massively parallel, multi-core GPUs = 1000s of Compute Cores
  • Infra CapEx
  • GPUs more expensive than CPUs
  • Underutilization of GPU resources (< 15%)
slide-35
SLIDE 35

35

AHV: Graphics by Day / Compute by Night

Display by Day

  • Graphics Mode
  • VDI
  • Design & Visualization

DAYTIME NIGHT

VDI/Visualization GPU/Compute Follow the Sun ! 100% GPU CapEx Utilization!

Compute by Night

  • CUDA Mode
  • Simulation, Processing
  • Analysis
  • Number Crunching
slide-36
SLIDE 36

36

  • Support for Nvidia’s cutting-edge Quadro RTX technology
  • Prism App for Nvidia Dashboard
  • vGPU live migration (Coming soon)

Future developments

slide-37
SLIDE 37

37

Conclusion: AHV drives GPU infrastructure elasticity

  • Fully Supports Nvidia GRID technology
  • Intuitive GPU management workflow
  • Built-in scale out technology
  • Storage scale out and elasticity
  • Data locality
  • GPU config simplicity (metrics, usage, alerts)
  • Storage scale out and elasticity
  • Infrastructure elasticity
  • 100% GPU utilization
  • CUDA/Graphics mode support
  • CapEx & TCO savings
  • Workload support
  • HPC
  • AI/ML (Inferencing, TensorFlow)
  • High End Design, Simulation
slide-38
SLIDE 38

DEMO – Multi-vGPU per VM

slide-39
SLIDE 39

DEMO – Graphics by Day & Compute by Night

slide-40
SLIDE 40

Thank you! Questions?

slide-41
SLIDE 41

41

Your Enterprise Cloud Platform