SILVER RIDGE GROUPHardware Control Plane Request a briefing
Protocols and integrations

Nine protocols in the box. A tenth without a release.

Production drivers, not plugin scaffolding. Each one is configured through maps in the admin interface, no code, no redeploy, no vendor engagement.

Four tiers converge on one pipeline

Discrete transitions, a coil flipping, a BGP session dropping, a binary HVAC point changing state, become events in the same stream as threshold breaches and statistical anomalies.

Tier 01
IT compute
Redfish
IPMI 2.0
Servers · blades · storage · BMCs · CDUs
Tier 02
GPU and HPC
DCGM
SLURM
GPU nodes · clusters · job queues
Tier 03
Network
gNMI / OpenConfig
SNMP v1/v2c
Routers · switches · optical transport
Tier 04
OT and facility
Modbus TCP
MQTT
BACnet/IP
UPS · PDUs · cooling · generators · HVAC
One telemetry pipeline, one alarm surface
Sample storage · event normalization · threshold and event rules · Z-score anomaly detection · health scoring · cascade suppression
IT tier · read and control

Redfish / DMTF

Rack servers, blade systems, storage, liquid cooling. Aligned to DMTF 2025.x, and re-exposed northbound as a compliant aggregated service root.

Capability
  • Topology discovery across Systems, Chassis, Power, Thermal, Managers, Storage and ThermalEquipment
  • Session token management with automatic refresh; pooled async HTTP with exponential backoff
  • Event subscription on the BMC with HMAC-SHA256 validation of inbound callbacks
  • Metrics: CPU health, watts, voltage, current, power factor, per-sensor temperature, fan RPM, UPS charge and runtime, PDU mains, storage IOPS, coolant supply and return, flow, pressure, pump speed
Verified against
HPE iLODell iDRACLenovo XCCSupermicroOpenBMCIntel RSD
IT tier · read and control

IPMI 2.0 / RMCP+

Out-of-band management for the fleet that predates Redfish, and it is always a larger share of the estate than the refresh plan assumes.

Capability
  • SDR sensor enumeration executed off the event loop
  • CPU, inlet, exhaust, memory and PCH temperatures; fan RPM; power and PSU draw; 12V, 5V, 3.3V, VCore and Vbat rails; current
  • Two-phase sensor naming: per-asset override map, then a vendor pattern library, then a slug fallback
  • Control: power on, off, reset, cycle and graceful shutdown through a named command map
Verified against
HPE iLO 4/5Dell iDRAC 7/8/9SupermicroLenovo XCCCisco CIMCFujitsu iRMCIBM FSPOracle ILOM
GPU tier · telemetry only

DCGM / dcgm-exporter

GPU cluster telemetry into the same alarm and health surface as the rest of the facility. No agent change on the node.

Capability
  • Scrapes any dcgm-exporter endpoint, Kubernetes DaemonSet, bare-metal node or standalone daemon
  • Eleven DCGM fields mapped into a canonical per-GPU namespace with automatic unit conversion
  • Per-node GPU index filter for selective ingestion
  • Uncorrectable ECC errors raise a P1 on any non-zero reading, regardless of workload state
Metrics per GPU
Junction temperaturePower drawSM utilisationMemory copy utilSM clockECC uncorrectableFramebuffer used / totalNVLink bandwidthPCIe TX / RX
Scheduler tier · telemetry only

SLURM / slurmrestd

The scheduler is infrastructure. Cluster assets are first class, and job-to-node assignment is a platform relationship the driver owns.

Capability
  • Polls the slurmrestd job API with a configurable partition filter per asset
  • Cluster telemetry: running jobs, pending jobs, allocated nodes, node utilisation
  • System-managed job-to-GPU-node relationships that operators cannot hand-edit, marked as such in the UI
  • Kubernetes and OpenStack Ironic adapters cover the same workload-context role for k8s and bare-metal estates
Why it matters

A GPU node under a running training job is hot, loud and fully utilised by design. Workload state suppresses the anomaly penalty so health scores stay meaningful, while hardware faults continue to fire at full priority.

Network tier · read and control

gNMI / OpenConfig

The streaming successor to SNMP for modern network gear, now production standard across every major platform.

Capability
  • Subscribe in ONCE mode with JSON_IETF or PROTO encoding, per asset
  • Per-asset path map: any OpenConfig YANG path to a platform metric with unit, scale and labels
  • Interface counters and errors, BGP session state and prefix counts, optical TX/RX power, OSNR, chromatic dispersion, laser bias, system CPU and memory
  • Named SetRequest commands for interface admin state and BGP operations; optional TLS verification
Platform support
Cisco IOS XRCisco NX-OSJuniper JunosArista EOSNokia SR-OS / SR-LinuxCiena Waveserver

Status flips on admin state, oper state and BGP sessions raise events the moment they happen, no polling overhead to detect them.

OT tier · read and control

Modbus TCP

The lingua franca of the facility floor. UPS, PDUs, cooling, generators, power meters and industrial PLCs from any vendor.

Capability
  • Batched holding and input register reads with contiguous-range merging to cut round trips
  • Five data types plus bit extraction from any bit of a 16-bit register
  • Scale, offset and metric labelling per register; JSON register maps with no code changes
  • Named writes across single and multiple coils and registers
  • Binary register transitions emit discrete status events into the alarm pipeline
Standards note

DMTF Redfish 2025.2 formally added Modbus target and connection-method modelling to the standard. This platform was already built that way, Redfish BMC aggregation and Modbus facility integration behind one northbound API.

OT tier · read and control

MQTT

Smart PDUs, environmental sensors, energy meters and BMS gateways, anything already publishing to a broker.

Capability
  • Persistent per-asset broker connection with single- and multi-level wildcard subscriptions
  • JSON path extraction per entry with scale, offset and metric labelling
  • Buffered ingestion drained on every scheduler tick, so bursts do not drop samples
  • Named publish commands with configurable QoS and retain flag
  • Credentials stored encrypted at rest
Note

Topics carrying pure discrete signals rather than physical units automatically emit status-change events on transition, so a contact closure is an event and not a number nobody reads.

Network and OT · telemetry only

SNMP v1/v2c

The long tail. Everything in the estate that will never speak anything newer, still on the same alarm surface as everything else.

Capability
  • Per-asset OID map with named metrics, scale, offset and unit
  • Version, timeout and retry count configurable per asset
  • Delivered as a versioned Protocol Pack and hot-registered at startup
  • Telemetry only, SNMP has no write path in this platform by design
Typical devices
Switches and routersUPSRack PDUsEnvironmental sensors
OT tier · read and control

BACnet/IP, ASHRAE 135

Past the server room door: chillers, air handling units, VAV boxes, cooling towers and the BMS gateways in front of them.

Capability
  • Seven object types: analog, binary and multi-state input, output and value, plus accumulator and pulse converter
  • Per-asset point map from object type and instance to a platform metric with scale and offset
  • Binary and multi-state transitions emit discrete status events
  • Named WriteProperty commands with BACnet priority, defaulting to manual operator override
  • Device instance addressing with optional network number for routed networks
Controller support
Any ASHRAE 135 controllerBMS gatewaysField controllersChiller plantsAHU / VAV
Protocol Packs

The tenth protocol is a file upload, not a roadmap item.

A pack bundles a driver, a JSON schema for its configuration and UI hints. Upload it, and the platform writes the source, imports it and registers it in the live protocol registry, no restart, no release train. The SNMP driver ships this way, which is how we know the path works in production.

Versioned

Every published version is immutable, and operators choose which is active per asset type.

Reversible

Roll back to a prior version from the admin interface when a map change misbehaves.

Governed

Pack upload is restricted to the platform-operator tier above all tenant accounts.

Candidates already scoped
OPC-UAPROFINETLONworksModbus RTU over TCP

Send us your device list. We will tell you what is covered today.

Request a briefing
Silver Ridge Group LLC · Irvine, California
+1 714-381-7883