GREEN
Energy-saving Toolkit for Large Scale Computing Resources
Contact
Michele Muggiri, valorisation@crs4.it
Challenge
The need to implement policies to contain energy consumption has an increasingly important role in industries and homes. For high performance computing (HPC) systems this needis even more acute: next to the canonic measure of the computing power in Flops/sec there is now the value of Watts/flop to measure the energy efficiency.
Newer operating systems come with features allowing the reduction of consumption, for example by reducing the clock of the unused CPUs/cores. However this feature is present mainly in desktop systems, while little tools exist in the HPC server-class operating systems, such as that of the CRS4 computer center.
Overview
GREEN is a toolkit for energy efficiency in high-performance and large-scale computing systems. Thanks to GREEN you can check the status of the computing cluster and intervene by switching off all the inactive nodes. GREEN, by means of its integration with the user job submission and control scheduler, keeps active only the actually used computers and turns them off when they are again inactive. GREEN is constituted by two main parts:
- a database that keeps the updated status of the computers and the related information such as scheduler used, on-off controls, ignition and shutdown policies and other information about hardware and software;
- an algorithmic module which controls and updates the database, select the machines on which to operate, and performs the start and stop operations.
Innovative features
The flexibility of GREEN derives from its modular architecture. The modularity allows its easy adaptation to any Linux cluster computing, using a queue manager. It is equipped with hardware that supports protocols such as ILO, ILOM, IPMI or WoL.
Potential users
Technologists working at computer and/or data centers.
Impact sectors
Computer and electronic systems - Management of data centers and high performance computing centers.
Other resources
- Antonio Concas, Luca Carta e Marco Moro. High Performance Computing and Network. Collana Seminari per la valorizzazione dei risultati della ricerca al CRS4, Cagliari, 8 Novembre 2012.