Sizing Guide

NetApp Project Neo v4 is composed of four microservices (API, Worker, Extractor, NER) plus a PostgreSQL database. Resource requirements depend on the volume of files being indexed and the processing throughput needed.

Deployment Profiles

Profile	Files	CPU	RAM	GPU	Storage
Small	<50K	4 vCPU	8 GB	Optional	20 GB
Medium	50K-500K	8 vCPU	16 GB	Recommended	50 GB
Large	500K-2M	16 vCPU	32 GB	Recommended	100 GB
Enterprise	2M+	32+ vCPU	64+ GB	Required	200+ GB

The above figures represent the total resources across all services. See the per-service breakdown below for how to distribute resources.

Per-Service Resource Breakdown

API Service

Lightweight HTTP service handling REST requests, authentication, and routing. Minimal resource requirements.

• CPU: 1-2 vCPU
• RAM: 1-2 GB
• GPU: Not required
• I/O: Low

Worker Service

Handles file discovery, SMB crawling, scheduling, and Microsoft Graph uploads. CPU and network intensive during crawl operations.

• CPU: 2-8 vCPU (scales with concurrent crawl jobs)
• RAM: 2-8 GB
• GPU: Not required
• I/O: High network throughput (SMB + Graph API)

Extractor Service

Performs document content extraction (PDF, Office, images). Heavily benefits from GPU acceleration for OCR and document layout analysis.

• CPU: 2-8 vCPU
• RAM: 4-16 GB
• GPU: Recommended (CUDA for NVIDIA, ROCm for AMD)
• I/O: Moderate

NER Service

Runs Named Entity Recognition models against extracted text. GPU-intensive for inference workloads.

• CPU: 2-4 vCPU
• RAM: 4-8 GB
• GPU: Recommended for production workloads
• I/O: Low

PostgreSQL Database

Primary data store for metadata, extracted content, search indexes, and ACLs. I/O intensive, especially during large crawl operations.

• CPU: 2-4 vCPU
• RAM: 4-16 GB
• GPU: Not required
• I/O: High (SSD strongly recommended)

Database Sizing

The database requirements depend on the size of your source data and the number of indexed documents:

• General rule: ~2 GB of database storage per 100 GB of source documents indexed (approximately 2% of source data size)
• With full-text content storage: Database size may increase to 5-10% of source data size
• NER results: Add approximately 10-20% to the base database size if NER is enabled

Ensure your PostgreSQL instance has adequate disk I/O performance. SSDs are strongly recommended for production deployments.

GPU Variants

Neo provides GPU-accelerated container variants for the Extractor and NER services:

• CUDA (NVIDIA): Images tagged with -cuda suffix
• ROCm (AMD): Images tagged with -rocm suffix

GPU acceleration significantly improves extraction and NER throughput. For deployments processing more than 50K files, GPU is recommended. For Enterprise deployments, GPU is required for acceptable processing times.

Performance Benchmarking

For detailed performance benchmarks and tuning guidance, see Performance.