The Cray Advanced Platform Monitoring and Control (CAPMC) service enables direct hardware control of nodes, compute blades, router modules, and liquid cooled chassis. CAPMC talks to BMCs via Redfish to control power, query status, and manage power capping on target components. These controls enable an administrator and 3rd party software to more intelligently manage state and system-wide power consumption.
Administrators can use the cray CLI for power operations from any system that
has HTTPS access to the
System Management Services.
Third party software can access the API directly. Refer to the CAPMC API documentation for detailed information about API options and features.
The cray capmc command (see --help) can be used to control power to
specific components by specifying the component NID, xname, or group.
xname_onxname_offxname_reinitget_xname_statusCAPMC power control assumes that all cabinets and PDUs have been plugged in, breakers are on, and PDU controllers, BMCs, and other embedded controller are on, available, and have been discovered. Components have their power controlled in a pre-defined order to properly handle requests of dependent components.
Important: It is recommended to use the Boot Orchestration Service (BOS) to boot (power On), shutdown, and reboot compute nodes.
CAPMC uses xnames to specify entire cabinets or specific components throughout
the system. By default, CAPMC controls power to only one component at a time. A
--recursive true option can be passed to CAPMC using the cray CLI. When the
--recursive true option is included in a request, all of the sub-components of
the target component are included in the power command.
By the cabinet naming convention, each cabinet in the system is assigned a unique number. Cabinet numbers can range from 0-9999 and contain from 1-4 digits only.
Although manufacturing typically follows a sequential cabinet numbering scheme:
x1000–x2999x3000–x4999x5000–x5999s0, allx1000, x3000, x5000x1000c7, x3500c0 (Air-cooled cabinets are always chassis 0)x1000c7s3, x3500c0s15 (U15)x1000c7s3b0n0, x3500c0s15b1n0x3200c0s9 (U9)x3200c0s9b0n0get_power_cap_capabilitiesget_power_capset_power_capCAPMC is capable of setting node power limits on all supported compute node hardware in both liquid cooled cabinets and air cooled cabinets. This functionality enables external software to establish an upper bound, or estimate a minimum bound, on the amount of power a system may consume. Separate CAPMC calls are required to power cap different compute node types as each compute node type has its own power capping capabilities.
NOTE: Power capping is not supported for liquid cooled chassis, switch modules, compute blades, and any non-compute nodes (NCNs) in air cooled cabinets.
See the CAPMC Deprecation Notice for more information
Use the interfaces from Power control and query by xname:
node_onnode_offnode_reinitget_node_statusUse the interfaces from Power control and query by xname:
group_ongroup_offgroup_reinitget_group_statusUse the System Monitoring Application (SMA) Grafana instance:
get_node_energyget_node_energy_statsget_node_energy_countersUse the System Monitoring Application (SMA) Grafana instance:
get_system_parametersget_system_powerget_system_power_detailsget_node_rulesemergency_power_offget_nid_map