The same core platform design can be used for small or large systems. The following sections list some of the differences.
Small and medium system scaling
The tested platform design architecture is ready for use with a small airport system (250 - 500 Cameras).
Other options to consider are:
- Deploy a setup with 4 active + 1 standby VxRail node.
- If fewer analytic cameras are needed, reduce the GPUs to one per node. Using fewer GPUs is preferred to using less capable cards and provides the best option for future expansion.
- Memory cannot be reduced for the nodes since VMS and CV management VMs can be run anywhere across the cluster.
- While it might be possible to reduce the CPU core counts for very small systems, it is recommended that validation be performed before finalizing the specification.
With this approach, it is possible to scale a small or medium platform by adding up to two A40 GPUs per host. Add additional hosts to the cluster if the environment must expand over time.
Large system scaling
A large system has some notable characteristic differences. For example, the management VMs carry less overall system overhead when distributed across a large cluster.
The following guidelines can be used when scaling up to a large system:
- Calculate the number of nodes required based on:
- Each node can have a max of 1200 Mb/s or 300 Cameras at 3.7 Mb/s.
- Each node can host two CV applications processing a max of 50 analytic cameras total.
- Allocate two additional nodes for all management VMs, such as Primary/Secondary databases.
- Add one or more standby nodes for HA operations.
Example: Building a 5000 camera system
Our system requirements consist of:
- Enough capacity to process 5000 cameras @3.7 Mb/s
- Resources capable of running CV applications processing video from 1000 real-time analytic cameras
|Number of processing nodes
Camera archive = 5000/300 = 17 nodes
Camera analytics = 1000/50 = 20 nodes
Number required = the maximum from above = 20 nodes
|Number of management nodes