Software

The REGALE architecture will be implemented as an end-to-end, ready-to-use software stack for extreme-scale HPC systems. REGALE partners bring in the project a complete toolset capable of supporting the implementation of the proposed architecture. Within REGALE the incoming tools will be promoted to modules of a complete system following an open architecture.

REGALE tools (open source)

ToolPartner
Role within REGALE
Strawman Architecture
Repository
OARUGASystem Managerhttps://github.com/oar-team/oar
DCDBLRZMonitor, Monitoring Datahttps://gitlab.lrz.de/dcdb/dcdb
EARBSCMonitor, Node Manager, Job Manager, Monitoring Datahttps://github.com/eas4dc/EAR
MelissaUGAWorkflow Enginehttps://gitlab.inria.fr/melissa
RYAXRYAXWorkflow Enginehttps://github.com/RyaxTech
ExamonE4/UNIBOMonitor, Monitoring Datahttps://github.com/EEESlab/examon
COUNTDOWNCINECA/UNIBOJob Managerhttps://github.com/EEESlab/countdown
Control PulpUNIBONode Managerhttps://github.com/pulp-platform/control-pulp
BeBiDaRYAXSystem Managerhttps://gitlab.inria.fr/mmercier/bebida

REGALE tools (closed source)

Closed source software used in the REGALE project are BEO (Bull Energy Optimizer) and BDPO (Bull Dynamic Power Optimizer). More information:

REGALE prototypes

PrototypeLink
Integration Scenario 1closed source
Integration Scenario 2different tools with their own repository
Integration Scenario 3https://gitlab.bsc.es/ear_team/cluster_sim/
Integration Scenario 4closed source
Integration Scenario 5to be updated
REGALE Libraryto be updated

REGALE pilots

PilotLink
Pilot 1closed source
Pilot 2https://gitlab.inria.fr/melissa
Pilot 3closed source
Pilot 4https://github.com/SCiO-systems/regale-ryax-modules
Pilot 5https://github.com/mgroupntua/carbumpercrashsimulation

Sophistication artifacts

ArtifactsLink
Minimizing resource wasteclosed source
Moldability for energy efficiencyhttps://github.com/cslab-ntua/oar3/tree/mold
Node level power controlsclosed source
Thermal and power control on a node levelclosed source
Application aware power-cappingclosed source
Data-aware resource allocationclosed source
Co-scheduling at the node levelhttps://github.com/cslab-ntua/oar3/tree/dev
Co-scheduling within multicore processing unitsIn the design phase
Elastic resource managementhttps://github.com/RyaxTech/bebida-optimization-service/
HazardNet: A Thermal Hazard Prediction Framework for Datacentershttps://github.com/MSKazemi/HazardNet

Integration of tools in REGALE