From the Fire Hose Series:
Benchmarking and Scalability in the Virtual Workspace
Ruben Spruijt CTO @ Atlantis Computing @rspruijt | Ruben@rspruijt.com Benny Tritsch Founder & Principal Consultant @ DrTritsch.com @drtritsch | gtc@drtritsch.com
the Virtual Workspace Ruben Spruijt Benny Tritsch CTO @ Atlantis - - PowerPoint PPT Presentation
From the Fire Hose Series: Benchmarking and Scalability in the Virtual Workspace Ruben Spruijt Benny Tritsch CTO @ Atlantis Computing Founder & Principal Consultant @ DrTritsch.com @rspruijt | Ruben@rspruijt.com @drtritsch |
Ruben Spruijt CTO @ Atlantis Computing @rspruijt | Ruben@rspruijt.com Benny Tritsch Founder & Principal Consultant @ DrTritsch.com @drtritsch | gtc@drtritsch.com
Ruben Spruijt CTO @ Atlantis Computing @rspruijt
Founder & Consultant @ DrTritsch.com @drtritsch
Access from anywhere - ‘work isn’t a place’ Data is centralized - no synchronization of large datasets High performance – leverage GPU, Flash, 10GbE Security and control - protect intellectual property Flexible - support BYO and COPE scenarios Disaster recovery – multi-site and multi-datacenter Reduce costs – centralized management Ergonomics – no heat and low noise under the desk
Construction / Engineering
High Performance Computing (HPC)
Medical Oil and Gas Film and Media Automotive Energy Hi-Tech Electronics Aerospace
“The modern virtual workspace is the engine of the digital workplace for today’s workforce!”
Ruben Spruijt – CTO - Atlantis Computing
, Unified Comm.
Designers
Power Users Knowledge Workers Task Workers
25M 200M 400M
Virtual Workstation Virtual Desktop/Apps
Virtual Desktops Bare Metal
Windows OS Hypervisor Hypervisor Hyper visor VM VM VM VM VM VM VM VM VM Hyper visor VM VM VM
Direct Map Software Graphics Shared GPU GPU Pass-thru GPU Virtualization
GPU graphics driver VDI vendor’s shared graphics framework GPU virtualization component GPU GPU GPU GPU GPU GPU GPUs User Session CPU (Emulating GPU)
Dedicated GPU (GPU Pass-thru) Shared GPU API Intercept Software Rendered Graphics
Designers
Power Users Knowledge Workers Task Workers
25M 200M 400M
“Modern applications and browsers benefit from a GPU which massively improves User
be the standard for every virtual desktop and workstation solution.”
“The biggest competition isn't Citrix vs VMware,
The goal is to make the Virtual Workspace better, cheaper and faster than a PC.”
Ruben Spruijt – CTO - Atlantis Computing
Physical Hardware (Compute, BIOS, GPUs) Virtualization Platform + GPU Assignment Network (DNS, AD) Operating System (Master VMs) Client Devices Session Broker Management UIs Portal Interface / Secure Access Storage VDI Infrastructure / Remoting Protocol Applications Workspace Management (Profiles, Policies, Monitoring) User Management / Role Assignment
NETWORK STORAGE COMPUTE
COMPLEXITY “Too complex to design, build, maintain” COST “It’s not cost effective - CAPEX and OPEX” PERFORMANCE “User Experience and Admin Experience is bad”
*) Atlantis RA: http://www2.atlantiscomputing.com/WC_RA_HyperScale-VDI-XenDesktop.html | VDI VM: 2vCPUs, 2GB RAM, 40GB
“User Experience is KING. GPU capabilities and blazing fast solid state storage is the standard for a modern virtual workspace!”
Ruben Spruijt – CTO - Atlantis Computing
GDI/GDI+ WPF DirectX/OpenGL Video/Audio Silverlight Flash
Bandwidth Latency Packet Loss
Speed
Data transfer rate of a network connection Delay; amount of time to traverse a system Discarding of data packets (in percent)
Remoting Protocol (RDP/RemoteFX, ICA/HDX, PCoIP, Blast, …)
NOTE: GPUs cannot compensate bottlenecks caused by network bandwidth, latency and packet loss
in terms of speed
rate expected in routine operation
loads in an environment
some considerable time
when the number of users is increased
“We want to know what the actual User Experience is, not for designers alone but also for power users and knowledge workers.”
TeamRGE
Selected host telemetry data points Remoting protocol features Remote session performance counters Client capabilities and media redirection Network settings Compression algorithms and media synchronization mechanisms End-to-end frame rates User interface response times
“Benchmarking never represents the real world – it only reflects selected aspects. And frames per second are not the most relevant quality criteria in remoting scenarios.”
Client Host
Test Controller RemoteUX Tracker
Network Guest Control Observe Archive
WAN Emulator
GPU
Tesla M60 Tesla M6 GRID K1 GRID K2 Quadro M6000 GPU Maxwell Maxwell Kepler Kepler Maxwell # GPU 2 1 4 2 1 GPU Type GM204 GM204 GK107 GK104 GM200 Clock speed 1126 MHz 1126 MHz 850 MHz 745 MHz 988 MHz CUDA cores – shading units 4096 (2048/GPU) 1536 768 (192/GPU) 3072 (1536/GPU) 3072 GFLOPS 4612 4612 3090 3090 6070 Memory size 16GB GDDR5 8GB GDDR5 16GB DDR3 8GB GDDR5 12GB GDDR5 Memory bus 256 bits/GPU 256 bits/GPU 128 bits/GPU 256 bits/GPU 384 bits H.264 1080p30 streams 36 18 8 12 Grid vGPU CCU 2/4/8/16/32 1/2/4/8/16 16 32 N/A Form Factor PCIe 3.0 Dual Slot (rack servers) MXM (blade
servers)
PCIe 3.0 Dual Slot (rack servers) PCIe 3.0 Dual Slot (rack servers) PCIe 3.0 Dual Slot (rack servers)
300W 100W 130W 225W Cooling Active/ passive Bare board Active/ passive Active/ passive Active
and M60, Citrix XenDesktop and Windows 8.1
troubleshooting when setting up the test environment
“Given the improvements that a GPU provides to a Virtual Desktop and Workstation it’s not a question of whether or not you’ll use GPUs but when you’ll use them.”
Shawn Bass – CTO EUC - VMware
Ruben Spruijt CTO @ Atlantis Computing @rspruijt
Founder & Consultant @ DrTritsch.com @drtritsch