Description
BEO (Bull Energy optimizer) is a tool for monitoring the power, energy, temperature and performance of the whole cluster infrastructure. Based on out-of-band monitoring through standard protocols (IPMI, SNMP, Redfish) and in-band monitoring (BDPO) and on a consolidated and distributed database, BEO can provide energy related insights without instrumentation. Connected with a system manager, BEO can provide detailed accounting metrics for a given job. BEO can be easily extended to support any additional hardware. BEO contains secured (OpenID connect, LDAP compatible) interfaces: CLI, Rest API (OpenAPI specification) and customizable web user interface intended for system administrators and end users.
Integration
BEO can support any IPMI/Redfish/SNMP compliant hardware for out-of-band data collection. The BEO agent, can be easily extended to act as a node manager, to implement in-band power capping features, encapsulating any RAPL or other low-level API. BEO can set limits on power, energy, or temperature metrics and receive notifications when thresholds are crossed. Integration with a system manager can help enforce power caps on compute nodes.
Sophistication
In REGALE, BEO will be integrated with the system manager, to enable setting power limits to any part of the infrastructure, using power capping on the compute nodes, but with a global view of the power draw of the infrastructure, including the interconnect and storage. In addition, the collected monitoring data/application profiles will be combined with ML methods to estimate the energy/power impact of power- constrained systems on applications and provide adapted power budgets, and to enable dynamic energy optimization of applications.