Home > Workload Solutions > Safety & Security > White Papers > Vix Vizion Imagus Video Analytics Validation with Dell EMC PowerEdge Server and NVIDIA-T4 GPU > Test Architecture
The test setup used Dell EMC PowerEdge R740xd hosts with NVIDIA GPUs and 10 GbE network interfaces running the Imagus facial recognition software. The database was tested in two different configurations. In the local attached storage test environment, the database was hosted locally on SSD attached storage. In the attached storage scenario, the database was shared and hosted on a Dell EMC VNXe-1600 with an ISCSI interface. Cameras were simulated using a camera simulator application running on a network connected server. Images were transferred on demand using Real Time Streaming Protocol (RTSP). Each image stream was sent at 30 frames/sec using 1080P format h.264 compression high quality (4 Mb/s).
The Single Host test scenario was configured as follows:
The number of CPUs and the amount of memory on the host VM needed for a given number of streams and VMs was also being evaluated. We determined the minimum configuration for a single stream and then increased the number of streams and the CPU and memory available to determine optimal configurations under a variable number of streams. We also tested the ability to use multiple VMs to scale out the performance by using a minimal configuration on a single VM and adding identical VMs to see how the performance was affected while using multiple physical GPUs configured to use GRID vGPU technology.
The test scenario was basically the same as the single host single VM test, with the addition of a second identically configured R740xd host to be used as a target system for High Availability (HA) testing of both planned migration and host failure HA scenario. The HA testing also included a R740xd host running VMWare vCenter-7.0 as a Management platform for the vSphere HA configuration. Both HA targets were sharing the same storage, so there was no storage failover tested and the failover time was minimized.