Mango AI Center
Mango AI Center (MAC) manages AI server fleets, GPUs, MangoBoost DPUs, RDMA NICs
and Ethernet fabrics. These 0.5.7 guides cover 0.5.7-rocm15220261006r1.
Use the version menu for older releases.
Highlights
- Fleet health, hardware inventory, software versions, alerts, and historical metrics
- NVIDIA GPU monitoring through DCGM and AMD GPU monitoring through ROCm
- MangoBoost, NVIDIA Mellanox, and Broadcom RDMA NIC inventory and bandwidth
- LLDP network topology and SONiC / Dell OS10 switch management
- PCIe device trees, GPU–NIC hop-distance matrices, and physical slot views
- SDK package operations and parallel MangoBoost DPU firmware updates
- Host logs, MAC service logs, and an administrator audit log
- Read-only LLMBoost serving visibility and MASS storage provisioning when connected
Readings depend on hardware, drivers and exporters. Missing readings differ from zero. Load or refresh failures show warnings; visible data may be stale.
Architecture
- The master runs the dashboard, API, metrics-bridge, Prometheus for metric history, and Loki for host logs.
- Agents run hardware/software exporters, sys-manager, mango-sdk-agent and bmc-agent. GPU exporters match the installed NVIDIA or AMD hardware.
- A shared node token protects registration and master–agent requests. Dashboard accounts use separate authentication and roles.
- Prometheus scrapes agents and metrics-bridge. Agents relay logs through the authenticated API to Loki. BMC consoles use Redfish; readings can use in-band IPMI.
See Deployment Modes for all-in-one, master/agent, and hybrid arrangements, and Environment for configuration and ports.
Find a Dashboard Page
Select a Node for node pages. Fleet views cover all registered agents.
| Dashboard group | Scope | Pages |
|---|---|---|
| Cluster | Fleet | Overview · Nodes |
| Fabric | Fleet | Topology · Switches |
| Service | Fleet | LLMBoost · MASS |
| Hardware | Selected node | Server · GPU · NIC · Storage |
| Monitoring | Selected node | Performance · Power & thermal · System log |
| Maintenance | Selected node | BMC · Software · Containers |
| MAC log | Fleet | Services · Audit log |
| Settings | Fleet | Alert rules · Users · Announcements · Updates (MAC stack / Mango SDK) |
Ctrl+K searches pages available to your role. BMC and host/service logs require
operator or admin; users, announcements, audit logs and power actions require
admin. Viewers can read monitoring and connected service dashboards.
Connected Services
LLMBoost shows serving services, models, engines, gateways and metrics discovered through Kubernetes, endpoints or agent host ports. Configure connections in MAC; manage serving lifecycle in LLMBoost.
MASS shows capacity, volumes, groups, alerts and drives from a MASS Manager. Operators can provision storage and prepare node attachments. Both dashboards need reachable external services; MAC does not deploy them.
Where to Go Next
- Getting Started — install, deploy, and authenticate.
- User Guide — inspect the fleet and use dashboard pages.
- CLI Reference — configure and control MAC services.
- Troubleshooting — diagnose installation and runtime issues.