DeepSearch does not require specific hardware or configuration - it’s primarily a CPU/RAM/disk-based workload.
Here is what we recommend as the minimum baseline:
12 CPU cores
32 GB of RAM
200 GB for an Enterprise Nucleus Server (SSD’s are recommended for the best performance.) - Disk space should be extended as needed.
The majority of disk space will be consumed by the Elastic Search / OpenSearch index.
Ultimately, we recommend monitoring the installation, and adding resources as required.
Please review the Elastic Search System requirements for configuration of the Elastic Search / OpenSearch cluster.
Installation of DeepSearch stack requires:
A working installation of an Enterprise Nucleus Server
An instance of Elastic Search or OpenSearch service. Please refer to Search Backend Configuration for more information.
The DeepSearch service stack is shipped in two ways as a Helm chart and as a Docker Compose stack to match the current shipment and distribution method of Enterprise Nucleus Server. Please refer to the following documentation for both Helm and Docker Compose installations.
Below you will find two sets of deployment documentation. The first corresponds to the latest General Availability (GA) release, whereas the latter to the latest Public Beta release for Early Access customers.
General Availability (GA)
Public Beta (For Early Access/Pre-Release customers only. Click here to apply to the Early Access/Pre-Release Program.)
2022.2.1 (General Availability)¶
Vulnerability fixes based on security analysis
Observability and telemetry performance improvements
Search backend performance improvements (more than 2x speed-up)
Add connection retry delay to the storage client
Replaces previous integrated implementation
Performance and scalability improvement
Removed discovery registration
All services now connect to the Storage service directly
HTTP support for scalability improvement
Added health and readiness checks
Support for receiving data from Omniverse Farm agents (required Omniverse Farm Agent 103.5 and greater)
OpenSearch backend: rely on orjson serializer instead of default json serializer (~x3 speed-up in serialization time)
Bug fix: Return images for non-KNN queries (OM-72745)
Various dependency updates based on security analysis
Docker-compose deployment: Single deployment for DeepSearch and NGSearch
Helm chart deployment: initial release