Lakeside Case Study – sizing NVIDIA GRID vGPU
I am very proud to share the results of the work I did with my friend and coworker Magnar Johnsen from FirstPoint. We have been developing for a long time a smart new way of analyzing data (iOPS, CPU, Memory, GPU, Latency..) and different pieces of software was evaluated and Lakeside Software was the product we decided to go with and both our companies Poppelgaard.com and Firstpoint who are a Lakeside partner so we are legit to make assessments for clients using Lakeside Software. Below article helps you if you are about to size a NVIDIA GRID vGPU solution using either Citrix XenServer or VMware vSphere. If you have been at NVIDIA GTC, Citrix Synergy, BriForum, E2EVC or seen me talking at a Citrix User Group you might have seen the results of the work we did, thats changing on how we think applications are impacting that requires a GPU.
Magnar and I go way back.
Magnar and I have a long past with remote graphics and when we put our minds together we create something that is beautiful. Last year at Citrix Synergy audience could see how people could virtualize 3d remote graphics from Virtual Reality solution using Oculus Rift “Facebook” from the cloud. This was showed for the first time for audience and Citrix loved the idea and redesigned their exhibit about HDX 3D Pro next to FrameHawk, and sorry FrameHawk we stole all the attention with everybody wants to try the VR solution. The cool thing was VR was cool, but me and Magnar worked months “hard” to analyze the data we find at Aibel and we did very successful assessment and we shared the findings at Brian Madden BriForum in May 2014, just after Citrix Synergy, and the audience could for the first time learn the lessons we experienced and we could see applications behavior is not “just” what you might think or expect it to be. Analytics is the key to understand your app. Lets talk about why…
Lets dig into the case study we did at Aibel
Aibel is an industrial pioneer with a history dating back more than a century. With around 5,500 employees worldwide the company is a leading supplier of engineering services related to oil, gas and renewable energy.
Aibel is a huge user of desktop virtualisation technology and is currently delivering virtual desktops to around 4,500 users. The challenge was to work out what to do with the remaining 1500 CAD/CAM workstations where graphics intensive modelling of 3D designs tied them to powerful physical workstations.
With projects occurring at the eight Aibel locations in Norway as well as at the Aibel offices in Thailand and Singapore the amount of data flowing across the WAN was increasing and becoming unwieldy. Large engineering models needed to be accessed on location and shared with engineers and designers in all Aibel office locations.
Virtualisation was seen as the solution to these issues but the challenge was how to virtualise graphics intensive workloads without impacting user experience or drive uneconomic datacentre specifications.
They teamed up with Thomas Poppelgaard, an independent expert in virtualisation and GPU technologies and together they started an initial survey of Aibel’s situation.
GPU Acceleration Approach
It became quickly clear that the only way to successfully virtualise the 3D graphics workstations in an economically viable fashion was to deploy GPU acceleration technology in the datacentre. NVIDIA’s GRID cards would allow dedicated and shared GPU accelerators to be placed in the datacentre to be used by virtual desktops offloading server CPUs and removing user experience impact for other virtual desktop users.
Having identified the technology approach the challenge now was to work out how to size the workload of these 1500 users and design a solution that would deliver user experience at least as good as their current physical desktop experience. In order to do that the team needed to work out:
- What graphics applications are in use today
- Who is using which applications and workstations
- What GPU processor power is being consumed today by application, user and workstationPoppelgaard.com and FirstPoint brought in Lakeside Software’s SysTrack to do this job. Using SysTrack’s granular data collection model they could model a complete picture of application and GPU workload across all the users and workstations in use at Aibel today.
Using SysTrack the team has been able to build a complete inventory of the existing application and workstation estate and user behaviour. Using these analytics and SysTrack technologies a detailed specification and design for the future NVIDIA GRID based desktop virtualisation infrastructure has been created. Aibel is able to trust this design
as it is completely based on observed data captured from their existing estate and modelled using industry leading technologies from Lakeside and NVIDIA to create the optimal solution for Aibel’s situation. Key in any design is to not over provision the solution and thereby inflate the cost of the solution. On the other hand under provisioning will lead to a poor user experience and potential project failure. SysTrack ensures the right data is used to make the right decision for the future estate.
The assessment was primarily executed to understand Aibel engineers CAD applications such as Aveva PDMS, Bentley Microstation, etc. Shown below is a summary of average GPU usage for 1500 physical machines which revealed surprisingly that internet browsers are very GPU intensive compared to other applications.
Another benefit of leveraging SysTrack was the SysTrack MarketPlace report that had been co-authored by Lakeside and NVIDIA which allowed the team to convert all the data collected into accurate sizing of the number of NVIDIA GRID cards by model number required to offload the GPU workload.
Citrix technology professional – CTP, and Microsoft Most Valuable Professional MVP, Thomas Poppelgaard provides professional services. Write to me on my email firstname.lastname@example.org or call on my cell +45 53540356