Technical Solutions Architect at Denali Advanced Integration
Vendor
2020-09-11T17:53:23Z
Sep 11, 2020
While both systems do a very good job within the nodal limits for scaling a compute environment - it is the networking layer that has the most impact on GPU sharing across a cluster. From a practical perspective, VxRail nodes will typically offer a discount over native Nutanix nodes. With Nutanix you also need servers, disk, networking and GPU's. VxRail will be more integrated with sized nodes, Dell, Cisco or other networking and often the VSphere 7 licenses can be integrated - achieving a lower net cost.
Testing for optimizations on either VSAN or Nutanix to see which favors a vGPU system would be something I would like to see quantified. My guess is they are network constrained and differences between VxRail and a build your own Nutanix environment will be negligible for the same network. Would be good to confirm if there is an optimal point - say 40Gbps vs 10Gbps or 100Gbps for the inter-GPU message passing.
I don't know the technical details, but if you want to work with the VMware hypervisor it would be best to select the VxRail platform, which is a part of Dell and VMware.
The Nutanix platform would be the option to select if you are thinking of changing the VMware hypervisor for the Nutanix Acropolis hypervisor, which, besides working very well, has no cost and could represent savings to the customer.
Unit Manager & Solutions Architect at a tech services company with 1,001-5,000 employees
Reseller
2020-09-08T09:13:51Z
Sep 8, 2020
Best option for vSphere 7 is the VxRail as it designed on VMware technology and built in integrates with vSphere, on a Nutanix node you will be running vSphere on top on the underlining Nutanix software.
Ok, I have no hands-on with vxRail. But we have vGPU running on Nutanix with ESXi (vsphere). I am not aware of something like vGPU sharing across nodes/hosts. You did not mention which solution you have in mind as you would be about to choose in the process.
We have Citrix XenDesktop running. It works really smoothly most of the time. We only encountered a lagging problem once every few weeks which we are yet to solve. Perhaps something running in the back of our net.
With Nutanix you would have a solid base to work with.
You did not mention what automation is to be used. You could use something like Nutanix Calm as an automation solution.
Nutanix Cloud Infrastructure (NCI) and VxRail compete in the hyper-converged infrastructure market. NCI seems to have an edge in features due to its broader cloud model support, while VxRail offers superior VMware integration.Features: NCI offers data locality, compression, and deduplication within a hyper-converged environment and supports on-premises, edge, and public clouds. VxRail integrates deeply with VMware, featuring orchestration, a single management console, and VMware's vSAN for...
"Nutanix node you will be running vSphere on top on the underlining Nutanix software" This description is incorrect.
There is no "layer" between the hardware and the OS which is ESXI hypervisor on both cases.
While both systems do a very good job within the nodal limits for scaling a compute environment - it is the networking layer that has the most impact on GPU sharing across a cluster. From a practical perspective, VxRail nodes will typically offer a discount over native Nutanix nodes. With Nutanix you also need servers, disk, networking and GPU's. VxRail will be more integrated with sized nodes, Dell, Cisco or other networking and often the VSphere 7 licenses can be integrated - achieving a lower net cost.
Testing for optimizations on either VSAN or Nutanix to see which favors a vGPU system would be something I would like to see quantified. My guess is they are network constrained and differences between VxRail and a build your own Nutanix environment will be negligible for the same network. Would be good to confirm if there is an optimal point - say 40Gbps vs 10Gbps or 100Gbps for the inter-GPU message passing.
I don't know the technical details, but if you want to work with the VMware hypervisor it would be best to select the VxRail platform, which is a part of Dell and VMware.
The Nutanix platform would be the option to select if you are thinking of changing the VMware hypervisor for the Nutanix Acropolis hypervisor, which, besides working very well, has no cost and could represent savings to the customer.
Best option for vSphere 7 is the VxRail as it designed on VMware technology and built in integrates with vSphere, on a Nutanix node you will be running vSphere on top on the underlining Nutanix software.
Ok, I have no hands-on with vxRail.
But we have vGPU running on Nutanix with ESXi (vsphere). I am not aware of something like vGPU sharing across nodes/hosts. You did not mention which solution you have in mind as you would be about to choose in the process.
We have Citrix XenDesktop running. It works really smoothly most of the time. We only encountered a lagging problem once every few weeks which we are yet to solve. Perhaps something running in the back of our net.
With Nutanix you would have a solid base to work with.
You did not mention what automation is to be used. You could use something like Nutanix Calm as an automation solution.
further reading: Citrix XenApp and XenDesktop on Nutanix
Dev-Test: Cloud Development Environment & Testing Solutions | Nutanix
Let me know which platform you have in mind to narrow down suggestions.