Cylc from the NCAS point of view (NCAS Experience with Rose/Cylc) Rosalyn Hatcher, Annette Osprey, NCAS-CMS (Grenville Lister) Joint final IS-ENES2 workshop on Workflow Solutions in Earth System Modelling and Meta-Data Generation during Experiments, Lisbon, Sept 2016
NCAS Rose/Cylc - Outline • NCAS-CMS – who we are • Workflow - Historical Overview • Rose/Cylc – Platforms – Suites – Management – Training – Future directions
NCAS Rose/Cylc – who we are NCAS Computational Modelling Services - http://cms.ncas.ac.uk
NCAS Rose/Cylc - Historical Code repositories trunk mirrored Institutional from MO/IPSL Commits/Extracts compute/storage Working copies Job configuration Job submission UMUI (jobs?) Local to PUMA COMPUTE Mobilis Polaris HPC MONSooN ARCHER (NOC) (Leeds) Wales POST PROCESSING MONSooN ARCHER JASMIN PP PP, RDF LOTUS, MASS cluster, SCI VMs JASMIN VM
UM software (pre vn9.0) PUMA UMUI Database of user jobs FCM Graphical job editor Code manager Compilation and build HPC OASIS Atmosphere model Reconfiguration Coupler Dynamical core Prepares initial model Physics state Diagnostics (STASH) [JULES and/or UKCA] Ocean model Output file tools Input file tools Local / Jasmin Data processing Prepare ancillary data Analysis and visualisation
NCAS Rose/Cylc - Historical • Discovery – ad hoc • Manage code – software engineering tools • Configure/reconfigure experiment – manual process • Manage job submission How this works is highly • Manage job failure/continuation dependent in individual users and frequently involves a good deal on • Manage output manual intervention. • Post process • Archive
NCAS Rose/Cylc - MOSRS M et O ffice S hared R epositories (UM, JULES, etc) Updated every 5 mins another site PUMA Local Mirror Local Mirror Make Make of Shared of Shared code changes code changes Repositories Repositories Extract code into Extract code into Rose suite Rose suite
NCAS Rose/Cylc Code repositories mirrored from Institutional MOSRS Commits/Extracts compute/storage Working copies Job configuration Rose/Cylc Job submission MOSRS Local to PUMA COMPUTE HPC Mobilis MONSooN Polaris ARCHER Wales POST PROCESSING MONSooN ARCHER JASMIN PP PP, RDF LOTUS, MASS cluster, SCI VMs JASMIN VM
UM software (vn10.0 onwards) PUMA MOSRS UMUI Rose Database of user jobs FCM Graphical job editor Cylc / Rose Code manager Job submission Workflow manager Compilation and build HPC OASIS Atmosphere model Reconfiguration Coupler Dynamical core Prepares initial model Physics state Diagnostics (STASH) [JULES and/or UKCA] Ocean model Output file tools Input file tools Local / Jasmin Data processing Prepare ancillary data Analysis and visualisation
NCAS Rose/Cylc - Platforms • PUMA (cylc daemons, polling) • ARCHER MONSooN (MO managed) – Service nodes Lander – RDF Analytics cluster 76 Rose VM Rose/Cylc Cylc VM • JASMIN users HPC – jasmin-xfer1 PP – jasmin-cylc (MO) • JASMIN-Reading (running jules locally) 119 Rose/Cylc users • Polaris (on the way)
NCAS Rose/Cylc - Suites Many suites! • Initial proliferation (support) • Greater convergence Standard Suites (?) • GA7 – ACSIS/FEBBRAIO • GC3 – HighresMIP • GO5 Suite development in production runs • NEMOVAR Many moving parts/points of failure • UKESM • Nesting
NCAS Rose/Cylc - Suites post processing postproc file conversion move from scratch to RDF remove from scratch checksum pptransfer pull files to jasmin checksum
NEMO data assimilation suite
NCAS Rose/Cylc - Training All but retired our UMUI-based training! • UMUI Conversion course (Sept 2016, Spring 2017) (http://cms.ncas.ac.uk/wiki/UmTraining/RoseSept2016) • 3-day UM Introduction (November and April) – Rose/Cylc based • 5-day UKCA Training (January 2016) – Rose/Cylc based (http://www.ukca.ac.uk/wiki/index.php/UKCA_Training_January_2016)
NCAS Rose/Cylc – Management … Management • Installation/testing of new releases and upgrades Consistency across platforms • UM, GCOM, other installation/testing – Rose-stem Still learning about the capabilities of the system Support • Porting suites • Troubleshooting/debugging • Increased level infrastructure support eg users writing their own guis
NCAS Rose/Cylc - Future • PUMA – central submission hub cloud-based PUMA (JASMIN VM)
NCAS Rose/Cylc - Future • PUMA – central submission hub cloud-based PUMA (JASMIN VM) • UM in the cloud – experiments in AWS ongoing Rose/Cylc control?
NCAS Rose/Cylc - Future • PUMA – central submission hub cloud-based PUMA (JASMIN VM) • UM in the cloud – experiments in AWS ongoing Rose/Cylc control? • MIP data workflow PP LOTUS MONSooN ARCHER VMs RDF -C work GWS MASS RDF ET
Recommend
More recommend