System Requirements

Deployment specifications.

PRISM deployment requirements — published transparently with measured benchmarks.
All figures reflect actual production configurations.

01 · Hardware Tiers

Three deployment tiers.

Sized for the most common production scenarios.
Baseline: 10 concurrent users running typical OLAP workloads on standard BI datasets.

Minimum
10 users

Concurrent (peak latency may rise) · operating threshold

vCPU8 cores
Memory16 GB
Disk500 GB SSD
Network1 Gbps
Recommended
10 users

Concurrent · response headroom · stable ops

vCPU16 cores
Memory32 GB
Disk500 GB NVMe
Network1 Gbps
Enterprise
100+ users

Company-wide · multi-department · scale after tuning

ConfigApp / DB / ES 3-stack split
App×2 hosts (Active-Active)
DBPostgreSQL+AGE Active-Standby
SearchElasticsearch ×2 nodes
Performance benchmark At 20 concurrent users on recommended hardware, OLAP queries returned p50 of 417 ms and p95 of 499 ms.
Memory footprint stays under 200 MB per query, regardless of dataset size.
02 · Air-Gapped Deployment

Three firewall paths. That's all.

PRISM deploys into fully isolated air-gapped networks.
Only three communication paths require firewall configuration — internal container traffic stays within the Docker network.

PathFromToPort / ProtocolPurpose
[1]User PCPRISM Server443 / TCP (HTTPS)Web Access
[2]PRISM ServerAI Inference ServerLLM API Port / TCPLLM Inference Call (or external SaaS = 443/TCP)
[3-8]PRISM ServerCustomer Source DBDB Port / TCPSource Data Query (Oracle 1521, MSSQL 1433, PG 5432, MySQL 3306, DB2 50000, SAP OData 443)
[9]Docker Internal Net · PostgreSQL · Elasticsearch · RedisNo firewall needed
What the customer prepares Only the ports in the table above need to be opened on the firewall. The AI inference server and load balancer are separate hosts. All communication stays within the customer's air-gapped network — PRISM does not send data out to the internet.
03 · High Availability

Optional redundancy for mission-critical deployments.

PRISM ships on a single host by default.
When non-stop availability is required, a reference HA architecture separates each tier — preserving service even if a single host fails.

TIER · 01

Application

Two app hosts in Active-Active configuration behind an L4 load balancer.
Supports 20 concurrent users with full failover.

TIER · 02

Database

PostgreSQL + AGE in Active-Standby with streaming replication.
RPO ≈ 0 · RTO in minutes.

What the customer prepares Approximately 6 servers plus a customer-side load balancer.
BI MATRIX handles installation, configuration, and ongoing support. Adopt only when operational requirements demand it.

Need a tailored deployment plan?

Share your concurrent user target, data sources, and compliance requirements.
We will return a sized deployment plan within 48 hours.