For the compute elements running the Symcloud Platform, the solution allows for the overlay of management services with compute and storage. See the table below.
Component | Configuration |
Control nodes | Minimum three PowerEdge R660 servers |
Worker nodes | Minimum four PowerEdge R760 servers with optional NVIDIA GPU |
CPU | Intel 4th Generation Xeon Scalable processors |
Networking | Minimum two PowerSwitch S5248F-ON |
Storage | Choice of either:
|
Software | Includes:
|
Kubernetes container platform | Symcloud Platform |
The control nodes are validated as PowerEdge R660 servers and the worker nodes as PowerEdge R760 servers. Symcloud Storage manages the NVMe or SSD storage disks. If GPU compute is required, the PowerEdge R760 should be deployed. Those platforms are better able to handle the additional power and cooling required for NVIDIA GPU accelerators. From a configuration perspective, dual-socket servers with at least 50 cores total are preferred. Typically, 32-core CPUs (64 total cores per node) are optimal. Memory configurations are configured starting at 512 GB.
For the storage elements, Dell PowerScale or Dell ECS should be deployed to handle the S3 storage and object store. The exact sizes and configurations of the storage will vary greatly from deployment to deployment based on both the size and types of information being stored. Cloud-native storage from Symcloud Platform is deployed across the compute nodes for use by the Spark, Kafka, TensorFlow, PyTorch, and other runtimes.
For the Dell PowerScale configuration, a three-node configuration with roughly 500 TB per node is recommended. For Dell ECS deployments, a five-node ECS EX500 configuration with 960 TB is the appropriate starting point, again based on the storage requirements.
The Dell PowerSwitch S5248F-ON is recommended for the networking component of the solution.